linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Logan Gunthorpe <logang@deltatee.com>
To: Jiasen Lin <linjiasen@hygon.cn>,
	linux-kernel@vger.kernel.org, linux-ntb@googlegroups.com,
	jdmason@kudzu.us
Cc: allenbh@gmail.com, dave.jiang@intel.com
Subject: Re: [PATCH] NTB: ntb_perf: Fix address err in perf_copy_chunk
Date: Mon, 11 Nov 2019 09:59:00 -0700	[thread overview]
Message-ID: <c340cb9e-9dcc-39f7-1e12-198ad2a47861@deltatee.com> (raw)
In-Reply-To: <8973e56c-ccce-2884-f4dc-4d0f8072a948@hygon.cn>



On 2019-11-11 12:51 a.m., Jiasen Lin wrote:
> 
> 
> On 2019/11/9 1:04, Logan Gunthorpe wrote:
>>
>>
>> On 2019-11-06 8:38 p.m., Jiasen Lin wrote:
>>> peer->outbuf is a virtual address which is get by ioremap, it can not
>>> be converted to a physical address by virt_to_page and page_to_phys.
>>> This conversion will result in DMA error, because the destination address
>>> which is converted by page_to_phys is invalid.
>>
>> Hmm, yes, ntb_perf is obviously wrong. I never noticed this, how did
>> this ever work?
>>
> 
> The default value of use_dma which is used to enable DMA engine to 
> measure NTB performance is zero, in other words, DMA engine is not used 
> by default. Thus, olny memcpy_toio is called in perf_copy_chunk and not 
> trigger this bug.
> 
> If we install driver with specified dma-enabled flag like this:
> insmod ntb_perf.ko use_dma=1, this issue will be triggered.

I've definitely tested with use_dma=1 in the past. But it looks like it
was broken by this problematic commit and I must have never personally
run the DMA tests after it:

5648e56d03fa ("NTB: ntb_perf: Add full multi-port NTB API support")

So it's probably worth adding a fixes tag to your patch.

>>> We Save the physical address in perf_setup_peer_mw, it is MMIO address
>>> of NTB BARx. Then fill the destination address of DMA descriptor with
>>> this physical address to guarantee that the address of memory write
>>> requests fall in memory window of NBT BARx.
>>
>> Using the physical address directly is also wrong and will not work in
>> the presence of an IOMMU. You should use dma_map_resource() for this.
>> See what was done in ntb_transport[1].
>>
> 
> Yes, my mistake. I will modify the code according to your suggestion and 
> test it on AMD and HYGON platforms with the IOMMU enabled. Maybe the 
> following patches are relied on, when IOMMU is enabled on AMD and HYGON 
> plartforms.
> 
> https://lore.kernel.org/patchwork/cover/1143155/
> https://lore.kernel.org/patchwork/patch/1143156/
> https://lore.kernel.org/patchwork/patch/1143157/

Thanks!

Logan

      reply	other threads:[~2019-11-11 16:59 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-11-07  3:38 [PATCH] NTB: ntb_perf: Fix address err in perf_copy_chunk Jiasen Lin
2019-11-08 17:04 ` Logan Gunthorpe
2019-11-11  7:51   ` Jiasen Lin
2019-11-11 16:59     ` Logan Gunthorpe [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=c340cb9e-9dcc-39f7-1e12-198ad2a47861@deltatee.com \
    --to=logang@deltatee.com \
    --cc=allenbh@gmail.com \
    --cc=dave.jiang@intel.com \
    --cc=jdmason@kudzu.us \
    --cc=linjiasen@hygon.cn \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-ntb@googlegroups.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).