2018-11-30 03:56:35

by Qian Cai

Subject: [PATCH] dma-debug: hns_enet_drv could use more DMA entries

The amount of DMA mappings from Hisilicon HNS ethernet devices is huge,
so it could trigger "DMA-API: debugging out of memory - disabling".

hnae_get_handle [1]
hnae_init_queue
hnae_init_ring
hnae_alloc_buffers [2]
debug_dma_map_page
dma_entry_alloc

[1] for (i = 0; i < handle->q_num; i++)
[2] for (i = 0; i < ring->desc_num; i++)

On this Huawei TaiShan 2280 aarch64 server, it has reached the limit
already,

4 (ports) x 16 (handles) x 1024 (rings) = 65536

Signed-off-by: Qian Cai <[email protected]>
---
kernel/dma/debug.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/kernel/dma/debug.c b/kernel/dma/debug.c
index 231ca4628062..ae91689cc9ad 100644
--- a/kernel/dma/debug.c
+++ b/kernel/dma/debug.c
@@ -43,8 +43,13 @@

/* allow architectures to override this if absolutely required */
#ifndef PREALLOC_DMA_DEBUG_ENTRIES
+/* amount of DMA mappings on this driver is huge. */
+#ifdef HNS_ENET
+#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 17)
+#else
#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
#endif
+#endif

enum {
dma_debug_single,
--
2.17.2 (Apple Git-113)

2018-11-30 08:44:19

by Christoph Hellwig

Subject: Re: [PATCH] dma-debug: hns_enet_drv could use more DMA entries

On Thu, Nov 29, 2018 at 10:54:56PM -0500, Qian Cai wrote:
> /* allow architectures to override this if absolutely required */
> #ifndef PREALLOC_DMA_DEBUG_ENTRIES
> +/* amount of DMA mappings on this driver is huge. */
> +#ifdef HNS_ENET
> +#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 17)
> +#else
> #define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
> #endif
> +#endif

How would this be defined in a header that leaks into this file?

I think we need to turn PREALLOC_DMA_DEBUG_ENTRIES into a user
selectable config options, as I really don't want to collect hacks like
this.

2018-11-30 09:17:25

by John Garry

Subject: Re: [PATCH] dma-debug: hns_enet_drv could use more DMA entries

+

Pasting original message at bottom.


On 30/11/2018 08:42, Christoph Hellwig wrote:
> On Thu, Nov 29, 2018 at 10:54:56PM -0500, Qian Cai wrote:
>> /* allow architectures to override this if absolutely required */
>> #ifndef PREALLOC_DMA_DEBUG_ENTRIES
>> +/* amount of DMA mappings on this driver is huge. */
>> +#ifdef HNS_ENET
>> +#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 17)
>> +#else
>> #define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
>> #endif
>> +#endif
>
> How would this be defined in a header that leaks into this file?
>
> I think we need to turn PREALLOC_DMA_DEBUG_ENTRIES into a user
> selectable config options, as I really don't want to collect hacks like
> this.
> _______________________________________________
> iommu mailing list
> [email protected]
> https://lists.linuxfoundation.org/mailman/listinfo/iommu
>
>







The amount of DMA mappings from Hisilicon HNS ethernet devices is huge,
so it could trigger "DMA-API: debugging out of memory - disabling".

hnae_get_handle [1]
hnae_init_queue
hnae_init_ring
hnae_alloc_buffers [2]
debug_dma_map_page
dma_entry_alloc

[1] for (i = 0; i < handle->q_num; i++)
[2] for (i = 0; i < ring->desc_num; i++)

On this Huawei TaiShan 2280 aarch64 server, it has reached the limit
already,

4 (ports) x 16 (handles) x 1024 (rings) = 65536

Signed-off-by: Qian Cai <[email protected]>
---
kernel/dma/debug.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/kernel/dma/debug.c b/kernel/dma/debug.c
index 231ca4628062..ae91689cc9ad 100644
--- a/kernel/dma/debug.c
+++ b/kernel/dma/debug.c
@@ -43,8 +43,13 @@

/* allow architectures to override this if absolutely required */
#ifndef PREALLOC_DMA_DEBUG_ENTRIES
+/* amount of DMA mappings on this driver is huge. */
+#ifdef HNS_ENET
+#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 17)
+#else
#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
#endif
+#endif

enum {
dma_debug_single,

2018-11-30 12:05:53

by Robin Murphy

Subject: Re: [PATCH] dma-debug: hns_enet_drv could use more DMA entries

On 30/11/2018 08:42, Christoph Hellwig wrote:
> On Thu, Nov 29, 2018 at 10:54:56PM -0500, Qian Cai wrote:
>> /* allow architectures to override this if absolutely required */
>> #ifndef PREALLOC_DMA_DEBUG_ENTRIES
>> +/* amount of DMA mappings on this driver is huge. */
>> +#ifdef HNS_ENET
>> +#define PREALLOC_DMA_DEBUG_ENTRIES (1 << 17)
>> +#else
>> #define PREALLOC_DMA_DEBUG_ENTRIES (1 << 16)
>> #endif
>> +#endif
>
> How would this be defined in a header that leaks into this file?

Apparently it wouldn't, because nothing defines that macro anywhere
anyway. CONFIG_HNS_ENET, on the other hand, would be available by virtue
of autoconf.h which is just about everywhere.

> I think we need to turn PREALLOC_DMA_DEBUG_ENTRIES into a user
> selectable config options, as I really don't want to collect hacks like
> this.

Agreed - we already have per-arch overrides scattered about, so growing
special cases in yet another place just makes even more of a mess.
Rounding all of it up into Kconfig would definitely make sense.

Robin.