archive mirror
 help / color / mirror / Atom feed
From: Robin Murphy <>
Subject: Re: dma_alloc_coherent not allocating memory from CMA Reserved
Date: Wed, 23 Sep 2020 14:53:26 +0100	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <20200923074314epcms5p8cfc65267615d27472df670f5806cb5a4@epcms5p8>

On 2020-09-23 08:43, Sathyavathi M wrote:
> Hi All,
> I am trying to allocate coherent memory for 33 MB in kerenl driver. and for that
> i have reserved CMA of 1024 MB, but from dmesg, i can see that address reserved
> for cma is different and what i get with dma_alloc_coherent is different. My pc
> is intel x86 machine and tried in different motherboard, but this issue is
> occuring in some specific PCs. please help me in debugging the actual issue,
> below are kernel bootup logs
> [ 0.014362] No NUMA configuration found
> [ 0.014363] Faking a node at [mem 0x0000000000000000-0x000000021edfffff]
> [ 0.014374] NODE_DATA(0) allocated [mem 0x21edd5000-0x21edfffff]
> [ 0.014538] *cma: Reserved 400 MiB at 0x0000000205c00000*
> [ 0.014541] Reserving 640MB of memory at 2512MB for crashkernel (System RAM: 8046MB)
> [ 0.014553] Zone ranges:
> [ 0.014554] DMA [mem 0x0000000000001000-0x0000000000ffffff]
> [ 0.014554] DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
> [ 0.014555] Normal [mem 0x0000000100000000-0x000000021edfffff]
> *at dma_alloc_coherent call
> [ 27.816062] dev->dma_33M_addr is f8000000*

I'm confused - if you got a DMA address back, then the allocation must 
have succeeded, so what exactly is the issue there? If the allocator 
managed to find a suitable amount of memory that your device can 
address, does it really matter exactly where it came from?

Also bear in mind that a DMA address is not necessarily the same as a 
physical address anyway, if for instance you have an IOMMU or other 
forms of interconnect translation. In fact the default behaviour of the 
intel-iommu driver for a PCI device is to allocate DMA addresses in 
naturally-aligned regions downwards from 4GB - if the first request in 
an empty address space was for 33MB, it would get rounded up to 64MB, 
wouldn't fit at 4GB - 64MB (0xfc000000) because that would clash with 
the IOAPIC region, so would end up at 4GB - 128MB (0xf8000000). Now 
where have I seen that before?...


> ---------------------------------------------------------------------------------------------------------------
> and below are the logs in working case, at driver dma_alloc_coherent api call we
> have address which is in range of what reserved for cma.
> at boot i get..
> Faking a node at [mem 0x0000000000000000-0x000000019fdfffff]
> NODE_DATA(0) allocated [mem 0x19fdd3000-0x19fdfdfff]
> *cma: Reserved 800 MiB at 0x000000016dc00000*
> Reserving 640MB of memory at 1792MB for crashkernel (System RAM: 6016MB)
> Zone ranges:
> DMA [mem 0x0000000000001000-0x0000000000ffffff]
> DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
> Normal [mem 0x0000000100000000-0x000000019fdfffff]
> *at dma_alloc_coherent call
> dev->dma_33M_addr is 16e200000*
> Please help me in solving this iissue, or can suggest any alternative way to
> allocate big coherent memory.
> Regards,
> Sathya
> _______________________________________________
> iommu mailing list
iommu mailing list

  reply	other threads:[~2020-09-23 13:53 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <CGME20200923074314epcms5p8cfc65267615d27472df670f5806cb5a4@epcms5p8>
2020-09-23  7:43 ` dma_alloc_coherent not allocating memory from CMA Reserved Sathyavathi M
2020-09-23 13:53   ` Robin Murphy [this message]
2020-09-23 14:05 ` Sathyavathi M
     [not found] <CGME20200918045919epcms5p170ffa89ef95533e98706f6bbdd744fc0@epcms5p1>
2020-09-18  4:59 ` ANKIT SONI

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:

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

  git send-email \ \ \ \ \

* 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).