powerpc/iommu: bypass DMA APIs for coherent allocations for pre-mapped memory

Leverage ARCH_HAS_DMA_MAP_DIRECT config option for coherent allocations as
well. This will bypass DMA ops for memory allocations that have been
pre-mapped.

Always set device bus_dma_limit when memory is pre-mapped. In some
architectures, like PowerPC, pmemory can be converted to regular memory via
daxctl command. This will gate the coherent allocations to pre-mapped RAM
only, by dma_coherent_ok().

Signed-off-by: Gaurav Batra <gbatra@linux.ibm.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20251107161105.85999-1-gbatra@linux.ibm.com

diff --git a/arch/powerpc/kernel/dma-iommu.c b/arch/powerpc/kernel/dma-iommu.c
index aa3689d619179d92a56ed831d6797f8750078eaf..73e10bd4d56db599b9956a4b5d286c8e4b8817f9 100644 (file)
--- a/arch/powerpc/kernel/dma-iommu.c
+++ b/arch/powerpc/kernel/dma-iommu.c
@@ -65,6 +65,21 @@ bool arch_dma_unmap_sg_direct(struct device *dev, struct scatterlist *sg,
 
        return true;
 }
+bool arch_dma_alloc_direct(struct device *dev)
+{
+       if (dev->dma_ops_bypass)
+               return true;
+
+       return false;
+}
+
+bool arch_dma_free_direct(struct device *dev, dma_addr_t dma_handle)
+{
+       if (!dev->dma_ops_bypass)
+               return false;
+
+       return is_direct_handle(dev, dma_handle);
+}
 #endif /* CONFIG_ARCH_HAS_DMA_MAP_DIRECT */
 
 /*
@@ -146,17 +161,12 @@ int dma_iommu_dma_supported(struct device *dev, u64 mask)
 
        if (dev_is_pci(dev) && dma_iommu_bypass_supported(dev, mask)) {
                /*
-                * dma_iommu_bypass_supported() sets dma_max when there is
-                * 1:1 mapping but it is somehow limited.
-                * ibm,pmemory is one example.
+                * fixed ops will be used for RAM. This is limited by
+                * bus_dma_limit which is set when RAM is pre-mapped.
                 */
-               dev->dma_ops_bypass = dev->bus_dma_limit == 0;
-               if (!dev->dma_ops_bypass)
-                       dev_warn(dev,
-                                "iommu: 64-bit OK but direct DMA is limited by %llx\n",
-                                dev->bus_dma_limit);
-               else
-                       dev_dbg(dev, "iommu: 64-bit OK, using fixed ops\n");
+               dev->dma_ops_bypass = true;
+               dev_info(dev, "iommu: 64-bit OK but direct DMA is limited by %llx\n",
+                        dev->bus_dma_limit);
                return 1;
        }
 

diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c
index eec333dd2e598c042f90b219e12f8cfd0a447fd7..5497b130e0263fa8405528c687418deee52cd471 100644 (file)
--- a/arch/powerpc/platforms/pseries/iommu.c
+++ b/arch/powerpc/platforms/pseries/iommu.c
@@ -1769,10 +1769,8 @@ out_failed:
 out_unlock:
        mutex_unlock(&dma_win_init_mutex);
 
-       /* If we have persistent memory and the window size is not big enough
-        * to directly map both RAM and vPMEM, then we need to set DMA limit.
-        */
-       if (pmem_present && direct_mapping && len != MAX_PHYSMEM_BITS)
+       /* For pre-mapped memory, set bus_dma_limit to the max RAM */
+       if (direct_mapping)
                dev->dev.bus_dma_limit = dev->dev.archdata.dma_offset +
                                                (1ULL << max_ram_len);
 

diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
index 4809204c674cc273e25d0e0a05a2a666a84095ee..8eff2f53fd86e93b121acc74a1796b38a70e84a3 100644 (file)
--- a/include/linux/dma-map-ops.h
+++ b/include/linux/dma-map-ops.h
@@ -395,11 +395,15 @@ bool arch_dma_map_sg_direct(struct device *dev, struct scatterlist *sg,
                int nents);
 bool arch_dma_unmap_sg_direct(struct device *dev, struct scatterlist *sg,
                int nents);
+bool arch_dma_alloc_direct(struct device *dev);
+bool arch_dma_free_direct(struct device *dev, dma_addr_t dma_handle);
 #else
 #define arch_dma_map_phys_direct(d, a)         (false)
 #define arch_dma_unmap_phys_direct(d, a)       (false)
 #define arch_dma_map_sg_direct(d, s, n)                (false)
 #define arch_dma_unmap_sg_direct(d, s, n)      (false)
+#define arch_dma_alloc_direct(d)            (false)
+#define arch_dma_free_direct(d, a)          (false)
 #endif
 
 #ifdef CONFIG_ARCH_HAS_SETUP_DMA_OPS

diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index 37163eb49f9faf45d771114c7fbfd0b5b9b64c4a..ee29c47781e333e0d1554fdfd3e051743228c3d9 100644 (file)
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c
@@ -638,7 +638,7 @@ void *dma_alloc_attrs(struct device *dev, size_t size, dma_addr_t *dma_handle,
        /* let the implementation decide on the zone to allocate from: */
        flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
 
-       if (dma_alloc_direct(dev, ops)) {
+       if (dma_alloc_direct(dev, ops) || arch_dma_alloc_direct(dev)) {
                cpu_addr = dma_direct_alloc(dev, size, dma_handle, flag, attrs);
        } else if (use_dma_iommu(dev)) {
                cpu_addr = iommu_dma_alloc(dev, size, dma_handle, flag, attrs);
@@ -679,7 +679,7 @@ void dma_free_attrs(struct device *dev, size_t size, void *cpu_addr,
                return;
 
        debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
-       if (dma_alloc_direct(dev, ops))
+       if (dma_alloc_direct(dev, ops) || arch_dma_free_direct(dev, dma_handle))
                dma_direct_free(dev, size, cpu_addr, dma_handle, attrs);
        else if (use_dma_iommu(dev))
                iommu_dma_free(dev, size, cpu_addr, dma_handle, attrs);