diff options
author | Benjamin Herrenschmidt <benh@kernel.crashing.org> | 2012-06-27 14:50:47 +1000 |
---|---|---|
committer | Anthony Liguori <aliguori@us.ibm.com> | 2012-06-27 16:33:26 -0500 |
commit | 7a0bac4da9c6a2e36d388412f3b4074b10429e8e (patch) | |
tree | 314c04cb25abc8f6144f2f9036f5efbd758c2864 | |
parent | edded454067bd9bda8a4a2a3ef54ae19474407da (diff) | |
download | hqemu-7a0bac4da9c6a2e36d388412f3b4074b10429e8e.zip hqemu-7a0bac4da9c6a2e36d388412f3b4074b10429e8e.tar.gz |
Add a memory barrier to DMA functions
The emulated devices can run simultaneously with the guest, so
we need to be careful with ordering of load and stores done by
them to the guest system memory, which need to be observed in
the right order by the guest operating system.
This adds a barrier call to the basic DMA read/write ops which
is currently implemented as a smp_mb(), but could be later
improved for more fine grained control of barriers.
Additionally, a _relaxed() variant of the accessors is provided
to easily convert devices who would be performance sensitive
and negatively impacted by the change.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
-rw-r--r-- | dma-helpers.c | 2 | ||||
-rw-r--r-- | dma.h | 53 |
2 files changed, 53 insertions, 2 deletions
diff --git a/dma-helpers.c b/dma-helpers.c index 2e09ceb..35cb500 100644 --- a/dma-helpers.c +++ b/dma-helpers.c @@ -31,6 +31,8 @@ static void do_dma_memory_set(dma_addr_t addr, uint8_t c, dma_addr_t len) int dma_memory_set(DMAContext *dma, dma_addr_t addr, uint8_t c, dma_addr_t len) { + dma_barrier(dma, DMA_DIRECTION_FROM_DEVICE); + if (dma_has_iommu(dma)) { return iommu_dma_memory_set(dma, addr, c, len); } @@ -13,6 +13,7 @@ #include <stdio.h> #include "hw/hw.h" #include "block.h" +#include "kvm.h" typedef struct DMAContext DMAContext; typedef struct ScatterGatherEntry ScatterGatherEntry; @@ -65,6 +66,31 @@ struct DMAContext { DMAUnmapFunc *unmap; }; +static inline void dma_barrier(DMAContext *dma, DMADirection dir) +{ + /* + * This is called before DMA read and write operations + * unless the _relaxed form is used and is responsible + * for providing some sane ordering of accesses vs + * concurrently running VCPUs. + * + * Users of map(), unmap() or lower level st/ld_* + * operations are responsible for providing their own + * ordering via barriers. + * + * This primitive implementation does a simple smp_mb() + * before each operation which provides pretty much full + * ordering. + * + * A smarter implementation can be devised if needed to + * use lighter barriers based on the direction of the + * transfer, the DMA context, etc... + */ + if (kvm_enabled()) { + smp_mb(); + } +} + static inline bool dma_has_iommu(DMAContext *dma) { return !!dma; @@ -88,8 +114,9 @@ static inline bool dma_memory_valid(DMAContext *dma, int iommu_dma_memory_rw(DMAContext *dma, dma_addr_t addr, void *buf, dma_addr_t len, DMADirection dir); -static inline int dma_memory_rw(DMAContext *dma, dma_addr_t addr, - void *buf, dma_addr_t len, DMADirection dir) +static inline int dma_memory_rw_relaxed(DMAContext *dma, dma_addr_t addr, + void *buf, dma_addr_t len, + DMADirection dir) { if (!dma_has_iommu(dma)) { /* Fast-path for no IOMMU */ @@ -101,6 +128,28 @@ static inline int dma_memory_rw(DMAContext *dma, dma_addr_t addr, } } +static inline int dma_memory_read_relaxed(DMAContext *dma, dma_addr_t addr, + void *buf, dma_addr_t len) +{ + return dma_memory_rw_relaxed(dma, addr, buf, len, DMA_DIRECTION_TO_DEVICE); +} + +static inline int dma_memory_write_relaxed(DMAContext *dma, dma_addr_t addr, + const void *buf, dma_addr_t len) +{ + return dma_memory_rw_relaxed(dma, addr, (void *)buf, len, + DMA_DIRECTION_FROM_DEVICE); +} + +static inline int dma_memory_rw(DMAContext *dma, dma_addr_t addr, + void *buf, dma_addr_t len, + DMADirection dir) +{ + dma_barrier(dma, dir); + + return dma_memory_rw_relaxed(dma, addr, buf, len, dir); +} + static inline int dma_memory_read(DMAContext *dma, dma_addr_t addr, void *buf, dma_addr_t len) { |