convert parisc to the generic dma-noncoherent code

convert parisc to the generic dma-noncoherent code

[Christoph Hellwig]

This should address all the comments raised last time.

[PATCH 1/3] parisc: merge pcx_dma_ops and pcxl_dma_ops

[Christoph Hellwig]

The only difference is that pcxl supports dma coherent allocations, while
pcx only supports non-consistent allocations and otherwise fails.

But dma_alloc* is not in the fast path, and merging these two allows an
easy migration path to the generic dma-noncoherent implementation, so
do it.

Signed-off-by: Christoph Hellwig <hch-jcswGhMUV9g@public.gmane.org>
---
 arch/parisc/include/asm/dma-mapping.h |  3 +-
 arch/parisc/kernel/pci-dma.c          | 80 ++++++++++++---------------
 arch/parisc/kernel/setup.c            |  8 +--
 arch/parisc/mm/init.c                 | 11 +---
 4 files changed, 43 insertions(+), 59 deletions(-)

diff --git a/arch/parisc/include/asm/dma-mapping.h b/arch/parisc/include/asm/dma-mapping.h
index 01e1fc057c83..eeec8dd18e74 100644
--- a/arch/parisc/include/asm/dma-mapping.h
+++ b/arch/parisc/include/asm/dma-mapping.h
@@ -22,8 +22,7 @@
 */
 
 #ifdef CONFIG_PA11
-extern const struct dma_map_ops pcxl_dma_ops;
-extern const struct dma_map_ops pcx_dma_ops;
+extern const struct dma_map_ops pa11_dma_ops;
 #endif
 
 extern const struct dma_map_ops *hppa_dma_ops;
diff --git a/arch/parisc/kernel/pci-dma.c b/arch/parisc/kernel/pci-dma.c
index 6df07ce4f3c2..d34cd6d34717 100644
--- a/arch/parisc/kernel/pci-dma.c
+++ b/arch/parisc/kernel/pci-dma.c
@@ -395,7 +395,7 @@ pcxl_dma_init(void)
 
 __initcall(pcxl_dma_init);
 
-static void *pa11_dma_alloc(struct device *dev, size_t size,
+static void *pcxl_dma_alloc(struct device *dev, size_t size,
 		dma_addr_t *dma_handle, gfp_t flag, unsigned long attrs)
 {
 	unsigned long vaddr;
@@ -422,16 +422,44 @@ static void *pa11_dma_alloc(struct device *dev, size_t size,
 	return (void *)vaddr;
 }
 
+static void *pcx_dma_alloc(struct device *dev, size_t size,
+		dma_addr_t *dma_handle, gfp_t flag, unsigned long attrs)
+{
+	void *addr;
+
+	if ((attrs & DMA_ATTR_NON_CONSISTENT) == 0)
+		return NULL;
+
+	addr = (void *)__get_free_pages(flag, get_order(size));
+	if (addr)
+		*dma_handle = (dma_addr_t)virt_to_phys(addr);
+
+	return addr;
+}
+
+static void *pa11_dma_alloc(struct device *dev, size_t size,
+		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
+{
+
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl)
+		return pcxl_dma_alloc(dev, size, dma_handle, gfp, attrs);
+	else
+		return pcx_dma_alloc(dev, size, dma_handle, gfp, attrs);
+}
+
 static void pa11_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
-	int order;
+	int order = get_order(size);
 
-	order = get_order(size);
-	size = 1 << (order + PAGE_SHIFT);
-	unmap_uncached_pages((unsigned long)vaddr, size);
-	pcxl_free_range((unsigned long)vaddr, size);
-	free_pages((unsigned long)__va(dma_handle), order);
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
+		size = 1 << (order + PAGE_SHIFT);
+		unmap_uncached_pages((unsigned long)vaddr, size);
+		pcxl_free_range((unsigned long)vaddr, size);
+
+		vaddr = __va(dma_handle);
+	}
+	free_pages((unsigned long)vaddr, get_order(size));
 }
 
 static dma_addr_t pa11_dma_map_page(struct device *dev, struct page *page,
@@ -560,7 +588,7 @@ static void pa11_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
 	flush_kernel_dcache_range((unsigned long)vaddr, size);
 }
 
-const struct dma_map_ops pcxl_dma_ops = {
+const struct dma_map_ops pa11_dma_ops = {
 	.alloc =		pa11_dma_alloc,
 	.free =			pa11_dma_free,
 	.map_page =		pa11_dma_map_page,
@@ -573,39 +601,3 @@ const struct dma_map_ops pcxl_dma_ops = {
 	.sync_sg_for_device =	pa11_dma_sync_sg_for_device,
 	.cache_sync =		pa11_dma_cache_sync,
 };
-
-static void *pcx_dma_alloc(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t flag, unsigned long attrs)
-{
-	void *addr;
-
-	if ((attrs & DMA_ATTR_NON_CONSISTENT) == 0)
-		return NULL;
-
-	addr = (void *)__get_free_pages(flag, get_order(size));
-	if (addr)
-		*dma_handle = (dma_addr_t)virt_to_phys(addr);
-
-	return addr;
-}
-
-static void pcx_dma_free(struct device *dev, size_t size, void *vaddr,
-		dma_addr_t iova, unsigned long attrs)
-{
-	free_pages((unsigned long)vaddr, get_order(size));
-	return;
-}
-
-const struct dma_map_ops pcx_dma_ops = {
-	.alloc =		pcx_dma_alloc,
-	.free =			pcx_dma_free,
-	.map_page =		pa11_dma_map_page,
-	.unmap_page =		pa11_dma_unmap_page,
-	.map_sg =		pa11_dma_map_sg,
-	.unmap_sg =		pa11_dma_unmap_sg,
-	.sync_single_for_cpu =	pa11_dma_sync_single_for_cpu,
-	.sync_single_for_device = pa11_dma_sync_single_for_device,
-	.sync_sg_for_cpu =	pa11_dma_sync_sg_for_cpu,
-	.sync_sg_for_device =	pa11_dma_sync_sg_for_device,
-	.cache_sync =		pa11_dma_cache_sync,
-};
diff --git a/arch/parisc/kernel/setup.c b/arch/parisc/kernel/setup.c
index 8d3a7b80ac42..5c8450a22255 100644
--- a/arch/parisc/kernel/setup.c
+++ b/arch/parisc/kernel/setup.c
@@ -97,14 +97,12 @@ void __init dma_ops_init(void)
 		panic(	"PA-RISC Linux currently only supports machines that conform to\n"
 			"the PA-RISC 1.1 or 2.0 architecture specification.\n");
 
-	case pcxs:
-	case pcxt:
-		hppa_dma_ops = &pcx_dma_ops;
-		break;
 	case pcxl2:
 		pa7300lc_init();
 	case pcxl: /* falls through */
-		hppa_dma_ops = &pcxl_dma_ops;
+	case pcxs:
+	case pcxt:
+		hppa_dma_ops = &pa11_dma_ops;
 		break;
 	default:
 		break;
diff --git a/arch/parisc/mm/init.c b/arch/parisc/mm/init.c
index 2607d2d33405..74842d28a7a1 100644
--- a/arch/parisc/mm/init.c
+++ b/arch/parisc/mm/init.c
@@ -19,7 +19,6 @@
 #include <linux/gfp.h>
 #include <linux/delay.h>
 #include <linux/init.h>
-#include <linux/pci.h>		/* for hppa_dma_ops and pcxl_dma_ops */
 #include <linux/initrd.h>
 #include <linux/swap.h>
 #include <linux/unistd.h>
@@ -616,17 +615,13 @@ void __init mem_init(void)
 	free_all_bootmem();
 
 #ifdef CONFIG_PA11
-	if (hppa_dma_ops == &pcxl_dma_ops) {
+	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
 		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 		parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 						+ PCXL_DMA_MAP_SIZE);
-	} else {
-		pcxl_dma_start = 0;
-		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
-	}
-#else
-	parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
+	} else
 #endif
+		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 
 	mem_init_print_info(NULL);
 
-- 
2.17.1

[PATCH 2/3] parisc: always use flush_kernel_dcache_range for DMA cache maintainance

[Christoph Hellwig]

Current the S/G list based DMA ops use flush_kernel_vmap_range which
contains a few UP optimizations, while the rest of the DMA operations
uses flush_kernel_dcache_range.  The single vs sg operations are supposed
to have the same effect, so they should use the same routines.  Use
the more conservation version for now, but if people more familiar with
parisc think the vmap version is generally fine for DMA we should switch
all interfaces over to it.

Signed-off-by: Christoph Hellwig <hch-jcswGhMUV9g@public.gmane.org>
---
 arch/parisc/kernel/pci-dma.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/parisc/kernel/pci-dma.c b/arch/parisc/kernel/pci-dma.c
index d34cd6d34717..1f85ca2c0c9e 100644
--- a/arch/parisc/kernel/pci-dma.c
+++ b/arch/parisc/kernel/pci-dma.c
@@ -537,7 +537,7 @@ static void pa11_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
 	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
 
 	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
+		flush_kernel_dcache_range(sg_virt(sg), sg->length);
 }
 
 static void pa11_dma_sync_single_for_cpu(struct device *dev,
@@ -568,7 +568,7 @@ static void pa11_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sgl
 	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
 
 	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
+		flush_kernel_dcache_range(sg_virt(sg), sg->length);
 }
 
 static void pa11_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sglist, int nents, enum dma_data_direction direction)
@@ -579,7 +579,7 @@ static void pa11_dma_sync_sg_for_device(struct device *dev, struct scatterlist *
 	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
 
 	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_vmap_range(sg_virt(sg), sg->length);
+		flush_kernel_dcache_range(sg_virt(sg), sg->length);
 }
 
 static void pa11_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
-- 
2.17.1

[PATCH 3/3] parisc: use generic dma_noncoherent_ops

[Christoph Hellwig]

Switch to the generic noncoherent direct mapping implementation.

Fix sync_single_for_cpu to do skip the cache flush unless the transfer
is to the device to match the more tested unmap_single path which should
have the same cache coherency implications.

Signed-off-by: Christoph Hellwig <hch-jcswGhMUV9g@public.gmane.org>
---
 arch/parisc/Kconfig                   |   4 +
 arch/parisc/include/asm/dma-mapping.h |   4 -
 arch/parisc/kernel/pci-dma.c          | 145 ++------------------------
 arch/parisc/kernel/setup.c            |   2 +-
 4 files changed, 16 insertions(+), 139 deletions(-)

diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig
index c480770fabcd..6fb2243a3b04 100644
--- a/arch/parisc/Kconfig
+++ b/arch/parisc/Kconfig
@@ -188,6 +188,10 @@ config PA20
 config PA11
 	def_bool y
 	depends on PA7000 || PA7100LC || PA7200 || PA7300LC
+	select ARCH_HAS_SYNC_DMA_FOR_CPU
+	select ARCH_HAS_SYNC_DMA_FOR_DEVICE
+	select DMA_NONCOHERENT_OPS
+	select DMA_NONCOHERENT_CACHE_SYNC
 
 config PREFETCH
 	def_bool y
diff --git a/arch/parisc/include/asm/dma-mapping.h b/arch/parisc/include/asm/dma-mapping.h
index eeec8dd18e74..44a9f97194aa 100644
--- a/arch/parisc/include/asm/dma-mapping.h
+++ b/arch/parisc/include/asm/dma-mapping.h
@@ -21,10 +21,6 @@
 ** flush/purge and allocate "regular" cacheable pages for everything.
 */
 
-#ifdef CONFIG_PA11
-extern const struct dma_map_ops pa11_dma_ops;
-#endif
-
 extern const struct dma_map_ops *hppa_dma_ops;
 
 static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)
diff --git a/arch/parisc/kernel/pci-dma.c b/arch/parisc/kernel/pci-dma.c
index 1f85ca2c0c9e..04c48f1ef3fb 100644
--- a/arch/parisc/kernel/pci-dma.c
+++ b/arch/parisc/kernel/pci-dma.c
@@ -21,13 +21,12 @@
 #include <linux/init.h>
 #include <linux/gfp.h>
 #include <linux/mm.h>
-#include <linux/pci.h>
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 #include <linux/string.h>
 #include <linux/types.h>
-#include <linux/scatterlist.h>
-#include <linux/export.h>
+#include <linux/dma-direct.h>
+#include <linux/dma-noncoherent.h>
 
 #include <asm/cacheflush.h>
 #include <asm/dma.h>    /* for DMA_CHUNK_SIZE */
@@ -437,7 +436,7 @@ static void *pcx_dma_alloc(struct device *dev, size_t size,
 	return addr;
 }
 
-static void *pa11_dma_alloc(struct device *dev, size_t size,
+void *arch_dma_alloc(struct device *dev, size_t size,
 		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
 {
 
@@ -447,7 +446,7 @@ static void *pa11_dma_alloc(struct device *dev, size_t size,
 		return pcx_dma_alloc(dev, size, dma_handle, gfp, attrs);
 }
 
-static void pa11_dma_free(struct device *dev, size_t size, void *vaddr,
+void arch_dma_free(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, unsigned long attrs)
 {
 	int order = get_order(size);
@@ -462,142 +461,20 @@ static void pa11_dma_free(struct device *dev, size_t size, void *vaddr,
 	free_pages((unsigned long)vaddr, get_order(size));
 }
 
-static dma_addr_t pa11_dma_map_page(struct device *dev, struct page *page,
-		unsigned long offset, size_t size,
-		enum dma_data_direction direction, unsigned long attrs)
+void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	void *addr = page_address(page) + offset;
-	BUG_ON(direction == DMA_NONE);
-
-	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC))
-		flush_kernel_dcache_range((unsigned long) addr, size);
-
-	return virt_to_phys(addr);
-}
-
-static void pa11_dma_unmap_page(struct device *dev, dma_addr_t dma_handle,
-		size_t size, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	BUG_ON(direction == DMA_NONE);
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	if (direction == DMA_TO_DEVICE)
-		return;
-
-	/*
-	 * For PCI_DMA_FROMDEVICE this flush is not necessary for the
-	 * simple map/unmap case. However, it IS necessary if if
-	 * pci_dma_sync_single_* has been called and the buffer reused.
-	 */
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle), size);
+	flush_kernel_dcache_range((unsigned long)phys_to_virt(paddr), size);
 }
 
-static int pa11_dma_map_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction direction,
-		unsigned long attrs)
+void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
+		size_t size, enum dma_data_direction dir)
 {
-	int i;
-	struct scatterlist *sg;
-
-	BUG_ON(direction == DMA_NONE);
-
-	for_each_sg(sglist, sg, nents, i) {
-		unsigned long vaddr = (unsigned long)sg_virt(sg);
-
-		sg_dma_address(sg) = (dma_addr_t) virt_to_phys(vaddr);
-		sg_dma_len(sg) = sg->length;
-
-		if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-			continue;
-
-		flush_kernel_dcache_range(vaddr, sg->length);
-	}
-	return nents;
+	flush_kernel_dcache_range((unsigned long)phys_to_virt(paddr), size);
 }
 
-static void pa11_dma_unmap_sg(struct device *dev, struct scatterlist *sglist,
-		int nents, enum dma_data_direction direction,
-		unsigned long attrs)
-{
-	int i;
-	struct scatterlist *sg;
-
-	BUG_ON(direction == DMA_NONE);
-
-	if (attrs & DMA_ATTR_SKIP_CPU_SYNC)
-		return;
-
-	if (direction == DMA_TO_DEVICE)
-		return;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_dcache_range(sg_virt(sg), sg->length);
-}
-
-static void pa11_dma_sync_single_for_cpu(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
-{
-	BUG_ON(direction == DMA_NONE);
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle),
-			size);
-}
-
-static void pa11_dma_sync_single_for_device(struct device *dev,
-		dma_addr_t dma_handle, size_t size,
-		enum dma_data_direction direction)
-{
-	BUG_ON(direction == DMA_NONE);
-
-	flush_kernel_dcache_range((unsigned long) phys_to_virt(dma_handle),
-			size);
-}
-
-static void pa11_dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sglist, int nents, enum dma_data_direction direction)
-{
-	int i;
-	struct scatterlist *sg;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_dcache_range(sg_virt(sg), sg->length);
-}
-
-static void pa11_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sglist, int nents, enum dma_data_direction direction)
-{
-	int i;
-	struct scatterlist *sg;
-
-	/* once we do combining we'll need to use phys_to_virt(sg_dma_address(sglist)) */
-
-	for_each_sg(sglist, sg, nents, i)
-		flush_kernel_dcache_range(sg_virt(sg), sg->length);
-}
-
-static void pa11_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
+void arch_dma_cache_sync(struct device *dev, void *vaddr, size_t size,
 	       enum dma_data_direction direction)
 {
 	flush_kernel_dcache_range((unsigned long)vaddr, size);
 }
-
-const struct dma_map_ops pa11_dma_ops = {
-	.alloc =		pa11_dma_alloc,
-	.free =			pa11_dma_free,
-	.map_page =		pa11_dma_map_page,
-	.unmap_page =		pa11_dma_unmap_page,
-	.map_sg =		pa11_dma_map_sg,
-	.unmap_sg =		pa11_dma_unmap_sg,
-	.sync_single_for_cpu =	pa11_dma_sync_single_for_cpu,
-	.sync_single_for_device = pa11_dma_sync_single_for_device,
-	.sync_sg_for_cpu =	pa11_dma_sync_sg_for_cpu,
-	.sync_sg_for_device =	pa11_dma_sync_sg_for_device,
-	.cache_sync =		pa11_dma_cache_sync,
-};
diff --git a/arch/parisc/kernel/setup.c b/arch/parisc/kernel/setup.c
index 5c8450a22255..4e87c35c22b7 100644
--- a/arch/parisc/kernel/setup.c
+++ b/arch/parisc/kernel/setup.c
@@ -102,7 +102,7 @@ void __init dma_ops_init(void)
 	case pcxl: /* falls through */
 	case pcxs:
 	case pcxt:
-		hppa_dma_ops = &pa11_dma_ops;
+		hppa_dma_ops = &dma_noncoherent_ops;
 		break;
 	default:
 		break;
-- 
2.17.1

Re: convert parisc to the generic dma-noncoherent code

[Christoph Hellwig]

ping?  Any comments?

On Tue, Jun 19, 2018 at 09:04:52AM +0200, Christoph Hellwig wrote:
> This should address all the comments raised last time.
> 
> 
> _______________________________________________
> iommu mailing list
> iommu-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
> https://lists.linuxfoundation.org/mailman/listinfo/iommu
---end quoted text---

Re: convert parisc to the generic dma-noncoherent code

[Helge Deller]

On 11.07.2018 17:34, Christoph Hellwig wrote:
> ping?  Any comments?

I applied those 3 patches on top of git head, and booted the
32-bit kernel on a HP 715/64 (PCX-L) and a HP B160L (PCX-L2).
On both machines I had problems with those drivers which use
DMA (I checked specifically the lasi NIC card, which is the 
main onboard NIC card in both machines).
Getting IP via DHCP was unreliable, pings on the same network 
to both machines gave lost packets, login via ssh sometimes failed
and so on.

So, there is definitively some cache-flush missing in this patchset.

I'm happy to test any further patches, but I'l be blocked at least
until end of next week. It would be good if James could do some 
testing on his PCX-machine too, since that machine is different
than mine.

Thanks for your efforts to clean that up!
Helge

> On Tue, Jun 19, 2018 at 09:04:52AM +0200, Christoph Hellwig wrote:
>> This should address all the comments raised last time.
For the record: Here is the link to the patchbundle:
https://patchwork.kernel.org/bundle/helge/DMA_OP/

Re: convert parisc to the generic dma-noncoherent code

[John David Anglin]

On 2018-07-13 4:14 AM, Helge Deller wrote:
> On 11.07.2018 17:34, Christoph Hellwig wrote:
>> ping?  Any comments?
> I applied those 3 patches on top of git head, and booted the
> 32-bit kernel on a HP 715/64 (PCX-L) and a HP B160L (PCX-L2).
> On both machines I had problems with those drivers which use
> DMA (I checked specifically the lasi NIC card, which is the
> main onboard NIC card in both machines).
> Getting IP via DHCP was unreliable, pings on the same network
> to both machines gave lost packets, login via ssh sometimes failed
> and so on.
>
> So, there is definitively some cache-flush missing in this patchset.
Possibly, the code should use flush_kernel_vmap_range() as it purges the 
TLB entries used
for the cache flush.  Some of the routines in pci-dma.c currently use it.

Dave

-- 
John David Anglin  dave.anglin@bell.net

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: convert parisc to the generic dma-noncoherent code

[Christoph Hellwig]

On Sun, Jul 15, 2018 at 12:29:37PM -0400, John David Anglin wrote:
> On 2018-07-13 4:14 AM, Helge Deller wrote:
>> On 11.07.2018 17:34, Christoph Hellwig wrote:
>>> ping?  Any comments?
>> I applied those 3 patches on top of git head, and booted the
>> 32-bit kernel on a HP 715/64 (PCX-L) and a HP B160L (PCX-L2).
>> On both machines I had problems with those drivers which use
>> DMA (I checked specifically the lasi NIC card, which is the
>> main onboard NIC card in both machines).
>> Getting IP via DHCP was unreliable, pings on the same network
>> to both machines gave lost packets, login via ssh sometimes failed
>> and so on.
>>
>> So, there is definitively some cache-flush missing in this patchset.
> Possibly, the code should use flush_kernel_vmap_range() as it purges the 
> TLB entries used
> for the cache flush.  Some of the routines in pci-dma.c currently use it.

That might be worth a try and is already discussed in my description.

Also it might be good to check that the last patch really is the culprit,
I'm pretty certain it is, but a double check never hurts.

Re: convert parisc to the generic dma-noncoherent code

[Helge Deller]

On 17.07.2018 15:19, Christoph Hellwig wrote:
> On Sun, Jul 15, 2018 at 12:29:37PM -0400, John David Anglin wrote:
>> On 2018-07-13 4:14 AM, Helge Deller wrote:
>>> On 11.07.2018 17:34, Christoph Hellwig wrote:
>>>> ping?  Any comments?
>>> I applied those 3 patches on top of git head, and booted the
>>> 32-bit kernel on a HP 715/64 (PCX-L) and a HP B160L (PCX-L2).
>>> On both machines I had problems with those drivers which use
>>> DMA (I checked specifically the lasi NIC card, which is the
>>> main onboard NIC card in both machines).
>>> Getting IP via DHCP was unreliable, pings on the same network
>>> to both machines gave lost packets, login via ssh sometimes failed
>>> and so on.
>>>
>>> So, there is definitively some cache-flush missing in this patchset.
>> Possibly, the code should use flush_kernel_vmap_range() as it purges the 
>> TLB entries used
>> for the cache flush.  Some of the routines in pci-dma.c currently use it.
> 
> That might be worth a try and is already discussed in my description.
> 
> Also it might be good to check that the last patch really is the culprit,
> I'm pretty certain it is, but a double check never hurts.

Somehow I now can't reproduce the cache issues any longer. Both, your original
patch and when using flush_kernel_vmap_range(), both work.

So, you can add a 
Tested-by: Helge Deller <deller@gmx.de>
to your original patch series.

Helge
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: convert parisc to the generic dma-noncoherent code

[Christoph Hellwig]

On Sun, Jul 22, 2018 at 05:45:40PM +0200, Helge Deller wrote:
> Somehow I now can't reproduce the cache issues any longer. Both, your original
> patch and when using flush_kernel_vmap_range(), both work.
> 
> So, you can add a 
> Tested-by: Helge Deller <deller-Mmb7MZpHnFY@public.gmane.org>
> to your original patch series.

I can merge this through the dma-mapping tree, but given that
it touches only arch/parisc/ it might make more sense to merge it
through the parisc tree.

Re: convert parisc to the generic dma-noncoherent code

[Helge Deller]

On 22.07.2018 21:57, Christoph Hellwig wrote:
> On Sun, Jul 22, 2018 at 05:45:40PM +0200, Helge Deller wrote:
>> Somehow I now can't reproduce the cache issues any longer. Both, your original
>> patch and when using flush_kernel_vmap_range(), both work.
>>
>> So, you can add a 
>> Tested-by: Helge Deller <deller-Mmb7MZpHnFY@public.gmane.org>
>> to your original patch series.
> 
> I can merge this through the dma-mapping tree, but given that
> it touches only arch/parisc/ it might make more sense to merge it
> through the parisc tree.

Sure, I will take it through the parisc tree.

Thanks!
Helge