Thread (10 messages) 10 messages, 2 authors, 2021-06-01
STALE1845d

[PATCH 1/8] MIPS: mm: XBurst CPU requires sync after DMA

From: Paul Cercueil <paul@crapouillou.net>
Date: 2021-05-30 17:18:26
Also in: linux-mips, lkml
Subsystem: mips, the rest · Maintainers: Thomas Bogendoerfer, Linus Torvalds

I am not sure why this is required, but if this is not enabled, reading
from a buffer in which data has been DMA'd may read incorrect values.

This used to happen for instance in mmc_app_send_scr()
(drivers/mmc/core/sd_ops.c), where data is DMA'd to a buffer then copied
by the CPU to a different location.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
---
 arch/mips/Kconfig              | 1 +
 arch/mips/mm/dma-noncoherent.c | 1 +
 2 files changed, 2 insertions(+)
diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index ed51970c08e7..310ce50ad285 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -112,6 +112,7 @@ config MACH_INGENIC
 	select SYS_SUPPORTS_LITTLE_ENDIAN
 	select SYS_SUPPORTS_ZBOOT
 	select DMA_NONCOHERENT
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
 	select IRQ_MIPS_CPU
 	select PINCTRL
 	select GPIOLIB
diff --git a/arch/mips/mm/dma-noncoherent.c b/arch/mips/mm/dma-noncoherent.c
index 212f3ce75a6b..3c4fc97b9f39 100644
--- a/arch/mips/mm/dma-noncoherent.c
+++ b/arch/mips/mm/dma-noncoherent.c
@@ -32,6 +32,7 @@ static inline bool cpu_needs_post_dma_flush(void)
 	case CPU_R12000:
 	case CPU_BMIPS5000:
 	case CPU_LOONGSON2EF:
+	case CPU_XBURST:
 		return true;
 	default:
 		/*
-- 
2.30.2
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help