在2023年12月30日十二月 下午7:38，Salvatore Bonaccorso写道：
> Hi,
>
> On Mon, Dec 11, 2023 at 07:22:24PM +0100, Greg Kroah-Hartman wrote:
>> 5.10-stable review patch. If anyone has any objections, please let me know.
>>
>> ------------------
>>
>> From: Jiaxun Yang <jiaxun.yang@xxxxxxxxxxx>
>>
>> commit edc0378eee00200a5bedf1bb9f00ad390e0d1bd4 upstream.
>>
>> There are some Loongson64 systems come with broken coherent DMA
>> support, firmware will set a bit in boot_param and pass nocoherentio
>> in cmdline.
>>
>> However nonconherent support was missed out when spin off Loongson-2EF
>> form Loongson64, and that boot_param change never made itself into
>> upstream.
>>
>> Support DMA noncoherent properly to get those systems working.
>>
>> Cc: stable@xxxxxxxxxxxxxxx
>> Fixes: 71e2f4dd5a65 ("MIPS: Fork loongson2ef from loongson64")
>> Signed-off-by: Jiaxun Yang <jiaxun.yang@xxxxxxxxxxx>
>> Signed-off-by: Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx>
>> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>> ---
>> arch/mips/Kconfig | 2 ++
>> arch/mips/include/asm/mach-loongson64/boot_param.h | 3 ++-
>> arch/mips/loongson64/env.c | 10 +++++++++-
>> 3 files changed, 13 insertions(+), 2 deletions(-)
>>
>> --- a/arch/mips/Kconfig
>> +++ b/arch/mips/Kconfig
>> @@ -468,6 +468,7 @@ config MACH_LOONGSON2EF
>>
>> config MACH_LOONGSON64
>> bool "Loongson 64-bit family of machines"
>> + select ARCH_DMA_DEFAULT_COHERENT
>> select ARCH_SPARSEMEM_ENABLE
>> select ARCH_MIGHT_HAVE_PC_PARPORT
>> select ARCH_MIGHT_HAVE_PC_SERIO
>> @@ -1379,6 +1380,7 @@ config CPU_LOONGSON64
>> select CPU_SUPPORTS_MSA
>> select CPU_DIEI_BROKEN if !LOONGSON3_ENHANCEMENT
>> select CPU_MIPSR2_IRQ_VI
>> + select DMA_NONCOHERENT
>> select WEAK_ORDERING
>> select WEAK_REORDERING_BEYOND_LLSC
>> select MIPS_ASID_BITS_VARIABLE
>> --- a/arch/mips/include/asm/mach-loongson64/boot_param.h
>> +++ b/arch/mips/include/asm/mach-loongson64/boot_param.h
>> @@ -117,7 +117,8 @@ struct irq_source_routing_table {
>> u64 pci_io_start_addr;
>> u64 pci_io_end_addr;
>> u64 pci_config_addr;
>> - u32 dma_mask_bits;
>> + u16 dma_mask_bits;
>> + u16 dma_noncoherent;
>> } __packed;
>>
>> struct interface_info {
>> --- a/arch/mips/loongson64/env.c
>> +++ b/arch/mips/loongson64/env.c
>> @@ -13,6 +13,8 @@
>> * Copyright (C) 2009 Lemote Inc.
>> * Author: Wu Zhangjin, wuzhangjin@xxxxxxxxx
>> */
>> +
>> +#include <linux/dma-map-ops.h>
>> #include <linux/export.h>
>> #include <linux/pci_ids.h>
>> #include <asm/bootinfo.h>
>> @@ -131,8 +133,14 @@ void __init prom_init_env(void)
>> loongson_sysconf.pci_io_base = eirq_source->pci_io_start_addr;
>> loongson_sysconf.dma_mask_bits = eirq_source->dma_mask_bits;
>> if (loongson_sysconf.dma_mask_bits < 32 ||
>> - loongson_sysconf.dma_mask_bits > 64)
>> + loongson_sysconf.dma_mask_bits > 64) {
>> loongson_sysconf.dma_mask_bits = 32;
>> + dma_default_coherent = true;
>> + } else {
>> + dma_default_coherent = !eirq_source->dma_noncoherent;
>> + }
>> +
>> + pr_info("Firmware: Coherent DMA: %s\n", dma_default_coherent ? "on" : "off");
>>
>> loongson_sysconf.restart_addr = boot_p->reset_system.ResetWarm;
>> loongson_sysconf.poweroff_addr = boot_p->reset_system.Shutdown;
>
> While preparing an update for Debian and finally building for all
> supported architecture, the builds for mipsel and mips64el were
> failing with:
>
> /<<PKGBUILDDIR>>/arch/mips/loongson64/env.c: In function
> 'prom_init_env':
> /<<PKGBUILDDIR>>/arch/mips/loongson64/env.c:138:3: error:
> 'dma_default_coherent' undeclared (first use in this function); did you
> mean 'dma_free_coherent'?
> 138 | dma_default_coherent = true;
> | ^~~~~~~~~~~~~~~~~~~~
> | dma_free_coherent
> /<<PKGBUILDDIR>>/arch/mips/loongson64/env.c:138:3: note: each
> undeclared identifier is reported only once for each function it
> appears in
> make[6]: *** [/<<PKGBUILDDIR>>/scripts/Makefile.build:291:
> arch/mips/loongson64/env.o] Error 1
> make[6]: *** Waiting for unfinished jobs....
>
> Is here a prerequisite missing for the commit?
>
> Backporting though 6d4e9a8efe3d ("driver core: lift dma_default_coherent into
> common code") which is from 5.12-rc1 though seems too intrusive, correct? Would
> the alternative be to just revert the 3ee7e2faef87 ("MIPS: Loongson64: Enable
> DMA noncoherent support") commit which landed in 5.10.204?
Hi,
Sorry for not spotting the issue in the first place.
It actually requires more dependencies such as c00a60d6f4a1 ("of: address: always
use dma_default_coherent for default coherency").
I'll try to produce a backport series but better to drop this patch from stable
for now.
Thanks
>
> Regards,
> Salvatore
--
- Jiaxun
