Good catch. But I hope c906/910 can directly use paddr here. It's
unnecessary to cause software translation & mmu translation here.
diff --git a/arch/riscv/mm/dma-noncoherent.c b/arch/riscv/mm/dma-noncoherent.c
index b0add983530a..30650a0c4481 100644
--- a/arch/riscv/mm/dma-noncoherent.c
+++ b/arch/riscv/mm/dma-noncoherent.c
@@ -24,13 +24,13 @@ void arch_sync_dma_for_device(phys_addr_t paddr,
size_t size,
switch (dir) {
case DMA_TO_DEVICE:
- ALT_CMO_OP(clean, vaddr, size, riscv_cbom_block_size);
+ ALT_CMO_OP(clean, vaddr, paddr, size, riscv_cbom_block_size);
break;
case DMA_FROM_DEVICE:
- ALT_CMO_OP(clean, vaddr, size, riscv_cbom_block_size);
+ ALT_CMO_OP(clean, vaddr, paddr, size, riscv_cbom_block_size);
break;
case DMA_BIDIRECTIONAL:
- ALT_CMO_OP(flush, vaddr, size, riscv_cbom_block_size);
+ ALT_CMO_OP(flush, vaddr, paddr, size, riscv_cbom_block_size);
break;
default:
break;
@@ -47,7 +47,7 @@ void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size,
break;
case DMA_FROM_DEVICE:
case DMA_BIDIRECTIONAL:
- ALT_CMO_OP(flush, vaddr, size, riscv_cbom_block_size);
+ ALT_CMO_OP(flush, vaddr, paddr, size, riscv_cbom_block_size);
break;
default:
break;
@@ -57,8 +57,9 @@ void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size,
void arch_dma_prep_coherent(struct page *page, size_t size)
{
void *flush_addr = page_address(page);
+ phys_addr_t paddr = PFN_PHYS(page_to_pfn(x));
- ALT_CMO_OP(flush, flush_addr, size, riscv_cbom_block_size);
+ ALT_CMO_OP(flush, flush_addr, paddr, size, riscv_cbom_block_size);
}
On Tue, Dec 27, 2022 at 10:03 AM Icenowy Zheng <[email protected]> wrote:
>
> The dcache.cva encoding shown in the comments are wrong, it's for
> dcache.cval1 (which is restricted to L1) instead.
>
> Fix this in the comment and in the hardcoded instruction.
>
> Signed-off-by: Icenowy Zheng <[email protected]>
> ---
> The code is tested on a LiteX SoC with OpenC906 core, and it
> successfully boots to Systemd on a SD card connected to LiteSDCard.
>
> This change should be not noticable on C906, but on multi-core C910
> cluster it should fixes something. Unfortunately TH1520 seems to be not
> so ready to test mainline patches on.
>
> arch/riscv/include/asm/errata_list.h | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/arch/riscv/include/asm/errata_list.h b/arch/riscv/include/asm/errata_list.h
> index 4180312d2a70..605800bd390e 100644
> --- a/arch/riscv/include/asm/errata_list.h
> +++ b/arch/riscv/include/asm/errata_list.h
> @@ -102,7 +102,7 @@ asm volatile(ALTERNATIVE( \
> * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> * 0000001 01001 rs1 000 00000 0001011
> * dcache.cva rs1 (clean, virtual address)
> - * 0000001 00100 rs1 000 00000 0001011
> + * 0000001 00101 rs1 000 00000 0001011
> *
> * dcache.cipa rs1 (clean then invalidate, physical address)
> * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> @@ -115,7 +115,7 @@ asm volatile(ALTERNATIVE( \
> * 0000000 11001 00000 000 00000 0001011
> */
> #define THEAD_inval_A0 ".long 0x0265000b"
> -#define THEAD_clean_A0 ".long 0x0245000b"
> +#define THEAD_clean_A0 ".long 0x0255000b"
> #define THEAD_flush_A0 ".long 0x0275000b"
> #define THEAD_SYNC_S ".long 0x0190000b"
>
> --
> 2.38.1
>
--
Best Regards
Guo Ren
在 2022-12-27星期二的 10:47 +0800，Guo Ren写道：
> Good catch. But I hope c906/910 can directly use paddr here. It's
> unnecessary to cause software translation & mmu translation here.
This could be an enhancement independent of this patchset.
In addition, I think this is some remaining of the design of the
original patchset, in which adds Zicbom support and then Xtheadcmo
support as some errata.
>
> diff --git a/arch/riscv/mm/dma-noncoherent.c b/arch/riscv/mm/dma-
> noncoherent.c
> index b0add983530a..30650a0c4481 100644
> --- a/arch/riscv/mm/dma-noncoherent.c
> +++ b/arch/riscv/mm/dma-noncoherent.c
> @@ -24,13 +24,13 @@ void arch_sync_dma_for_device(phys_addr_t paddr,
> size_t size,
>
> switch (dir) {
> case DMA_TO_DEVICE:
> - ALT_CMO_OP(clean, vaddr, size,
> riscv_cbom_block_size);
> + ALT_CMO_OP(clean, vaddr, paddr, size,
> riscv_cbom_block_size);
Maybe the macro should be renamed as ALT_CMO_OP_VPA or similar thing,
that means both VA and PA are provided and the most efficient one will
be chosen, because CMO operations with PA is still some T-Head-only
thing (Zicbom only supports VA).
> break;
> case DMA_FROM_DEVICE:
> - ALT_CMO_OP(clean, vaddr, size,
> riscv_cbom_block_size);
> + ALT_CMO_OP(clean, vaddr, paddr, size,
> riscv_cbom_block_size);
> break;
> case DMA_BIDIRECTIONAL:
> - ALT_CMO_OP(flush, vaddr, size,
> riscv_cbom_block_size);
> + ALT_CMO_OP(flush, vaddr, paddr, size,
> riscv_cbom_block_size);
> break;
> default:
> break;
> @@ -47,7 +47,7 @@ void arch_sync_dma_for_cpu(phys_addr_t paddr,
> size_t size,
> break;
> case DMA_FROM_DEVICE:
> case DMA_BIDIRECTIONAL:
> - ALT_CMO_OP(flush, vaddr, size,
> riscv_cbom_block_size);
> + ALT_CMO_OP(flush, vaddr, paddr, size,
> riscv_cbom_block_size);
> break;
> default:
> break;
> @@ -57,8 +57,9 @@ void arch_sync_dma_for_cpu(phys_addr_t paddr,
> size_t size,
> void arch_dma_prep_coherent(struct page *page, size_t size)
> {
> void *flush_addr = page_address(page);
> + phys_addr_t paddr = PFN_PHYS(page_to_pfn(x));
>
> - ALT_CMO_OP(flush, flush_addr, size, riscv_cbom_block_size);
> + ALT_CMO_OP(flush, flush_addr, paddr, size,
> riscv_cbom_block_size);
> }
>
> On Tue, Dec 27, 2022 at 10:03 AM Icenowy Zheng <[email protected]>
> wrote:
> >
> > The dcache.cva encoding shown in the comments are wrong, it's for
> > dcache.cval1 (which is restricted to L1) instead.
> >
> > Fix this in the comment and in the hardcoded instruction.
> >
> > Signed-off-by: Icenowy Zheng <[email protected]>
> > ---
> > The code is tested on a LiteX SoC with OpenC906 core, and it
> > successfully boots to Systemd on a SD card connected to LiteSDCard.
> >
> > This change should be not noticable on C906, but on multi-core C910
> > cluster it should fixes something. Unfortunately TH1520 seems to be
> > not
> > so ready to test mainline patches on.
> >
> > arch/riscv/include/asm/errata_list.h | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/riscv/include/asm/errata_list.h
> > b/arch/riscv/include/asm/errata_list.h
> > index 4180312d2a70..605800bd390e 100644
> > --- a/arch/riscv/include/asm/errata_list.h
> > +++ b/arch/riscv/include/asm/errata_list.h
> > @@ -102,7 +102,7 @@ asm
> > volatile(ALTERNATIVE( \
> > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > * 0000001 01001 rs1 000 00000 0001011
> > * dcache.cva rs1 (clean, virtual address)
> > - * 0000001 00100 rs1 000 00000 0001011
> > + * 0000001 00101 rs1 000 00000 0001011
> > *
> > * dcache.cipa rs1 (clean then invalidate, physical address)
> > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > @@ -115,7 +115,7 @@ asm
> > volatile(ALTERNATIVE( \
> > * 0000000 11001 00000 000 00000 0001011
> > */
> > #define THEAD_inval_A0 ".long 0x0265000b"
> > -#define THEAD_clean_A0 ".long 0x0245000b"
> > +#define THEAD_clean_A0 ".long 0x0255000b"
> > #define THEAD_flush_A0 ".long 0x0275000b"
> > #define THEAD_SYNC_S ".long 0x0190000b"
> >
> > --
> > 2.38.1
> >
>
>
Hi Icenowy,
I can't find this patch on lore or patchwork, (although the replies made
it). Would you please resend it once the discussion has died down?
Thanks,
Conor.
> > The dcache.cva encoding shown in the comments are wrong, it's for
> > dcache.cval1 (which is restricted to L1) instead.
> >
> > Fix this in the comment and in the hardcoded instruction.
> >
> > Signed-off-by: Icenowy Zheng <[email protected]>
> > ---
> > The code is tested on a LiteX SoC with OpenC906 core, and it
> > successfully boots to Systemd on a SD card connected to LiteSDCard.
> >
> > This change should be not noticable on C906, but on multi-core C910
> > cluster it should fixes something. Unfortunately TH1520 seems to be not
> > so ready to test mainline patches on.
> >
> > arch/riscv/include/asm/errata_list.h | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/riscv/include/asm/errata_list.h b/arch/riscv/include/asm/errata_list.h
> > index 4180312d2a70..605800bd390e 100644
> > --- a/arch/riscv/include/asm/errata_list.h
> > +++ b/arch/riscv/include/asm/errata_list.h
> > @@ -102,7 +102,7 @@ asm volatile(ALTERNATIVE( \
> > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > * 0000001 01001 rs1 000 00000 0001011
> > * dcache.cva rs1 (clean, virtual address)
> > - * 0000001 00100 rs1 000 00000 0001011
> > + * 0000001 00101 rs1 000 00000 0001011
> > *
> > * dcache.cipa rs1 (clean then invalidate, physical address)
> > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > @@ -115,7 +115,7 @@ asm volatile(ALTERNATIVE( \
> > * 0000000 11001 00000 000 00000 0001011
> > */
> > #define THEAD_inval_A0 ".long 0x0265000b"
> > -#define THEAD_clean_A0 ".long 0x0245000b"
> > +#define THEAD_clean_A0 ".long 0x0255000b"
> > #define THEAD_flush_A0 ".long 0x0275000b"
> > #define THEAD_SYNC_S ".long 0x0190000b"
Nice catch ! I experimented with T-Head RVB-ICE board on the up-to-date
upstream kernel, using device tree and some other bits from the vendor
kernel [1]. Without this patch, emmc on this board does not work since
noncoherent dma sync operations use incorrect 'clean' instruction.
Applying this patch fixes emmc operations, so
Tested-by: Sergey Matyukevich <[email protected]>
[1] https://github.com/T-head-Semi/linux/tree/linux-5.10.4
Regards,
Sergey
在 2022-12-31星期六的 01:12 +0300，Sergey Matyukevich写道：
> > > The dcache.cva encoding shown in the comments are wrong, it's for
> > > dcache.cval1 (which is restricted to L1) instead.
> > >
> > > Fix this in the comment and in the hardcoded instruction.
> > >
> > > Signed-off-by: Icenowy Zheng <[email protected]>
> > > ---
> > > The code is tested on a LiteX SoC with OpenC906 core, and it
> > > successfully boots to Systemd on a SD card connected to
> > > LiteSDCard.
> > >
> > > This change should be not noticable on C906, but on multi-core
> > > C910
> > > cluster it should fixes something. Unfortunately TH1520 seems to
> > > be not
> > > so ready to test mainline patches on.
> > >
> > > arch/riscv/include/asm/errata_list.h | 4 ++--
> > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/arch/riscv/include/asm/errata_list.h
> > > b/arch/riscv/include/asm/errata_list.h
> > > index 4180312d2a70..605800bd390e 100644
> > > --- a/arch/riscv/include/asm/errata_list.h
> > > +++ b/arch/riscv/include/asm/errata_list.h
> > > @@ -102,7 +102,7 @@ asm
> > > volatile(ALTERNATIVE( \
> > > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > > * 0000001 01001 rs1 000 00000 0001011
> > > * dcache.cva rs1 (clean, virtual address)
> > > - * 0000001 00100 rs1 000 00000 0001011
> > > + * 0000001 00101 rs1 000 00000 0001011
> > > *
> > > * dcache.cipa rs1 (clean then invalidate, physical address)
> > > * | 31 - 25 | 24 - 20 | 19 - 15 | 14 - 12 | 11 - 7 | 6 - 0 |
> > > @@ -115,7 +115,7 @@ asm
> > > volatile(ALTERNATIVE( \
> > > * 0000000 11001 00000 000 00000 0001011
> > > */
> > > #define THEAD_inval_A0 ".long 0x0265000b"
> > > -#define THEAD_clean_A0 ".long 0x0245000b"
> > > +#define THEAD_clean_A0 ".long 0x0255000b"
> > > #define THEAD_flush_A0 ".long 0x0275000b"
> > > #define THEAD_SYNC_S ".long 0x0190000b"
>
> Nice catch ! I experimented with T-Head RVB-ICE board on the up-to-
> date
> upstream kernel, using device tree and some other bits from the
> vendor
> kernel [1]. Without this patch, emmc on this board does not work
> since
> noncoherent dma sync operations use incorrect 'clean' instruction.
> Applying this patch fixes emmc operations, so
Thanks for testing on C910! I don't think this patch will make a
significant difference on C906, so it's valuable!
>
> Tested-by: Sergey Matyukevich <[email protected]>
>
> [1] https://github.com/T-head-Semi/linux/tree/linux-5.10.4
>
> Regards,
> Sergey