* Re: [dpdk-dev] [PATCH v3] bus/pci: resolve multiple NICs address conflicts
2019-11-12 2:22 ` [dpdk-dev] [PATCH v3] bus/pci: resolve multiple NICs address conflicts Wangyu (Eric)
@ 2019-11-12 7:01 ` Gavin Hu (Arm Technology China)
2019-11-12 12:29 ` Burakov, Anatoly
1 sibling, 0 replies; 3+ messages in thread
From: Gavin Hu (Arm Technology China) @ 2019-11-12 7:01 UTC (permalink / raw)
To: Wangyu (Eric), dev@dpdk.org
Cc: ferruh.yigit@intel.com, Linuxarm, humin (Q), dengxiaofeng,
Liyuan (Larry), nd
Hi Eric,
> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Wangyu (Eric)
> Sent: Tuesday, November 12, 2019 10:23 AM
> To: dev@dpdk.org
> Cc: ferruh.yigit@intel.com; Linuxarm <linuxarm@huawei.com>; humin (Q)
> <humin29@huawei.com>; dengxiaofeng <dengxiaofeng@huawei.com>;
> Liyuan (Larry) <Larry.T@huawei.com>
> Subject: [dpdk-dev] [PATCH v3] bus/pci: resolve multiple NICs address conflicts
Change the headline to "bus/pci: fix mapping issue when page size is greater than bar"?
> NIC address conflicts on 64K pagesize when using multiple NICs,
>
> as system will mmap 64K pagesize for NIC,
>
> but dev->mem_resource[i].len is 16K.
I would like the problem is described as follows and followed by analysis and fix.
When integrating NICs that request memory spaces less than a page size, the problem hits as follows:
" Cannot mmap device resource file %s to address: %p "
The root cause is the PCI memory space is mapped at a page size granule, without considering the page size to calculate the next address for future mapping is wrong.
The fix is to calculate the next address for future mappings using the real mapping size.
/Gavin
>
> Signed-off-by: beard-627 <dengxiaofeng@huawei.com>
> Signed-off-by: Wangyu (Eric) <seven.wangyu@huawei.com>
> Acked-by: Wei Hu <xavier.huwei@huawei.com>
> Acked-by: Min Hu <humin29@huawei.com>
> ---
> drivers/bus/pci/linux/pci_uio.c | 2 ++ drivers/bus/pci/linux/pci_vfio.c | 3 +++
> 2 files changed, 5 insertions(+)
>
> diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
> index 6dca05a..097dc19 100644
> --- a/drivers/bus/pci/linux/pci_uio.c
> +++ b/drivers/bus/pci/linux/pci_uio.c
> @@ -351,6 +351,8 @@
> pci_map_addr = RTE_PTR_ADD(mapaddr,
> (size_t)dev->mem_resource[res_idx].len);
>
> + pci_map_addr = RTE_PTR_ALIGN(pci_map_addr,
> sysconf(_SC_PAGE_SIZE));
Here "pci_map_addr" is, in effect, exactly the next_addr we talked about in V2, for use of next mapping, I vote for this fix.
/Gavin
> maps[map_idx].phaddr = dev->mem_resource[res_idx].phys_addr;
> maps[map_idx].size = dev->mem_resource[res_idx].len;
> maps[map_idx].addr = mapaddr;
> diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c
> index b8faa23..64cd84a 100644
> --- a/drivers/bus/pci/linux/pci_vfio.c
> +++ b/drivers/bus/pci/linux/pci_vfio.c
> @@ -750,6 +750,9 @@
> bar_addr = pci_map_addr;
> pci_map_addr = RTE_PTR_ADD(bar_addr, (size_t) reg->size);
>
> + pci_map_addr = RTE_PTR_ALIGN(pci_map_addr,
> + sysconf(_SC_PAGE_SIZE));
> +
> maps[i].addr = bar_addr;
> maps[i].offset = reg->offset;
> maps[i].size = reg->size;
> --
> 1.8.3.1
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [dpdk-dev] [PATCH v3] bus/pci: resolve multiple NICs address conflicts
2019-11-12 2:22 ` [dpdk-dev] [PATCH v3] bus/pci: resolve multiple NICs address conflicts Wangyu (Eric)
2019-11-12 7:01 ` Gavin Hu (Arm Technology China)
@ 2019-11-12 12:29 ` Burakov, Anatoly
1 sibling, 0 replies; 3+ messages in thread
From: Burakov, Anatoly @ 2019-11-12 12:29 UTC (permalink / raw)
To: Wangyu (Eric), dev@dpdk.org
Cc: ferruh.yigit@intel.com, Linuxarm, humin (Q), dengxiaofeng,
Liyuan (Larry)
On 12-Nov-19 2:22 AM, Wangyu (Eric) wrote:
> NIC address conflicts on 64K pagesize when using multiple NICs,
>
> as system will mmap 64K pagesize for NIC,
>
> but dev->mem_resource[i].len is 16K.
Suggested rewording:
===
bus/pci: align next mapping address on page boundary
Currently, the next address picked by PCI mapping infrastructure may be
page-unaligned due to BAR length being smaller than page size. This
leads to a situation where the requested map address is invalid,
resulting in mmap() call returning an arbitrary address, which will
later interfere with device BAR mapping in secondary processes.
Fix it by always aligning the next requested address on page boundary.
===
Also, probably missing Fixes: tag and Cc: stable.
>
> Signed-off-by: beard-627 <dengxiaofeng@huawei.com>
That's an unconventional name :)
> Signed-off-by: Wangyu (Eric) <seven.wangyu@huawei.com>
> Acked-by: Wei Hu <xavier.huwei@huawei.com>
> Acked-by: Min Hu <humin29@huawei.com>
> ---
> drivers/bus/pci/linux/pci_uio.c | 2 ++ drivers/bus/pci/linux/pci_vfio.c | 3 +++
> 2 files changed, 5 insertions(+)
>
> diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c index 6dca05a..097dc19 100644
> --- a/drivers/bus/pci/linux/pci_uio.c
> +++ b/drivers/bus/pci/linux/pci_uio.c
> @@ -351,6 +351,8 @@
> pci_map_addr = RTE_PTR_ADD(mapaddr,
> (size_t)dev->mem_resource[res_idx].len);
>
> + pci_map_addr = RTE_PTR_ALIGN(pci_map_addr, sysconf(_SC_PAGE_SIZE));
> +
> maps[map_idx].phaddr = dev->mem_resource[res_idx].phys_addr;
> maps[map_idx].size = dev->mem_resource[res_idx].len;
> maps[map_idx].addr = mapaddr;
> diff --git a/drivers/bus/pci/linux/pci_vfio.c b/drivers/bus/pci/linux/pci_vfio.c
> index b8faa23..64cd84a 100644
> --- a/drivers/bus/pci/linux/pci_vfio.c
> +++ b/drivers/bus/pci/linux/pci_vfio.c
> @@ -750,6 +750,9 @@
> bar_addr = pci_map_addr;
> pci_map_addr = RTE_PTR_ADD(bar_addr, (size_t) reg->size);
>
> + pci_map_addr = RTE_PTR_ALIGN(pci_map_addr,
> + sysconf(_SC_PAGE_SIZE));
> +
> maps[i].addr = bar_addr;
> maps[i].offset = reg->offset;
> maps[i].size = reg->size;
> --
> 1.8.3.1
>
>
As for contents of this patch (as long as the above is addressed):
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
--
Thanks,
Anatoly
^ permalink raw reply [flat|nested] 3+ messages in thread