linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Dongli Zhang <dongli.zhang@oracle.com>
To: Joe Jin <joe.jin@oracle.com>,
	iommu@lists.linux-foundation.org, linux-kernel@vger.kernel.org
Cc: konrad.wilk@oracle.com, hch@lst.de, m.szyprowski@samsung.com,
	robin.murphy@arm.com
Subject: Re: [PATCH RFC 1/1] swiotlb: add debugfs to track swiotlb buffer usage
Date: Fri, 7 Dec 2018 13:49:40 +0800	[thread overview]
Message-ID: <28a9f44a-d1bf-fb34-5a57-cfdb7bb23163@oracle.com> (raw)
In-Reply-To: <da16ecce-4e25-d246-3e0d-389e8e9e413e@oracle.com>



On 12/07/2018 12:12 AM, Joe Jin wrote:
> Hi Dongli,
> 
> Maybe move d_swiotlb_usage declare into swiotlb_create_debugfs():

I assume the call of swiotlb_tbl_map_single() might be frequent in some
situations, e.g., when 'swiotlb=force'.

That's why I declare the d_swiotlb_usage out of any functions and use "if
(unlikely(!d_swiotlb_usage))".

I think "if (unlikely(!d_swiotlb_usage))" incur less performance overhead than
calling swiotlb_create_debugfs() every time to confirm if debugfs is created. I
would declare d_swiotlb_usage statically inside swiotlb_create_debugfs() if the
performance overhead is acceptable (it is trivial indeed).


That is the reason I tag the patch with RFC because I am not sure if the
on-demand creation of debugfs is fine with maintainers/reviewers. If swiotlb
pages are never allocated, we would not be able to see the debugfs entry.

I would prefer to limit the modification within swiotlb and to not taint any
other files.

The drawback is there is no place to create or delete the debugfs entry because
swiotlb buffer could be initialized and uninitialized at very early stage.

> 
> void swiotlb_create_debugfs(void)
> {
> #ifdef CONFIG_DEBUG_FS
> 	static struct dentry *d_swiotlb_usage = NULL;
> 
> 	if (d_swiotlb_usage)
> 		return;
> 
> 	d_swiotlb_usage = debugfs_create_dir("swiotlb", NULL);
> 
> 	if (!d_swiotlb_usage)
> 		return;
> 
> 	debugfs_create_file("usage", 0600, d_swiotlb_usage,
> 			    NULL, &swiotlb_usage_fops);
> #endif
> }
> 
> And for io_tlb_used, possible add a check at the begin of swiotlb_tbl_map_single(),
> if there were not any free slots or not enough slots, return fail directly?

This would optimize the slots allocation path. I will follow this in next
version after I got more suggestions and confirmations from maintainers.


Thank you very much!

Dongli Zhang

> 
> Thanks,
> Joe
> On 12/5/18 7:59 PM, Dongli Zhang wrote:
>> The device driver will not be able to do dma operations once swiotlb buffer
>> is full, either because the driver is using so many IO TLB blocks inflight,
>> or because there is memory leak issue in device driver. To export the
>> swiotlb buffer usage via debugfs would help the user estimate the size of
>> swiotlb buffer to pre-allocate or analyze device driver memory leak issue.
>>
>> As the swiotlb can be initialized at very early stage when debugfs cannot
>> register successfully, this patch creates the debugfs entry on demand.
>>
>> Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
>> ---
>>  kernel/dma/swiotlb.c | 57 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 57 insertions(+)
>>
>> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
>> index 045930e..d3c8aa4 100644
>> --- a/kernel/dma/swiotlb.c
>> +++ b/kernel/dma/swiotlb.c
>> @@ -35,6 +35,9 @@
>>  #include <linux/scatterlist.h>
>>  #include <linux/mem_encrypt.h>
>>  #include <linux/set_memory.h>
>> +#ifdef CONFIG_DEBUG_FS
>> +#include <linux/debugfs.h>
>> +#endif
>>  
>>  #include <asm/io.h>
>>  #include <asm/dma.h>
>> @@ -73,6 +76,13 @@ static phys_addr_t io_tlb_start, io_tlb_end;
>>   */
>>  static unsigned long io_tlb_nslabs;
>>  
>> +#ifdef CONFIG_DEBUG_FS
>> +/*
>> + * The number of used IO TLB block
>> + */
>> +static unsigned long io_tlb_used;
>> +#endif
>> +
>>  /*
>>   * This is a free list describing the number of free entries available from
>>   * each index
>> @@ -100,6 +110,41 @@ static DEFINE_SPINLOCK(io_tlb_lock);
>>  
>>  static int late_alloc;
>>  
>> +#ifdef CONFIG_DEBUG_FS
>> +
>> +static struct dentry *d_swiotlb_usage;
>> +
>> +static int swiotlb_usage_show(struct seq_file *m, void *v)
>> +{
>> +	seq_printf(m, "%lu\n%lu\n", io_tlb_used, io_tlb_nslabs);
>> +	return 0;
>> +}
>> +
>> +static int swiotlb_usage_open(struct inode *inode, struct file *filp)
>> +{
>> +	return single_open(filp, swiotlb_usage_show, NULL);
>> +}
>> +
>> +static const struct file_operations swiotlb_usage_fops = {
>> +	.open           = swiotlb_usage_open,
>> +	.read           = seq_read,
>> +	.llseek         = seq_lseek,
>> +	.release        = single_release,
>> +};
>> +
>> +void swiotlb_create_debugfs(void)
>> +{
>> +	d_swiotlb_usage = debugfs_create_dir("swiotlb", NULL);
>> +
>> +	if (!d_swiotlb_usage)
>> +		return;
>> +
>> +	debugfs_create_file("usage", 0600, d_swiotlb_usage,
>> +			    NULL, &swiotlb_usage_fops);
>> +}
>> +
>> +#endif
>> +
>>  static int __init
>>  setup_io_tlb_npages(char *str)
>>  {
>> @@ -449,6 +494,11 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
>>  		pr_warn_once("%s is active and system is using DMA bounce buffers\n",
>>  			     sme_active() ? "SME" : "SEV");
>>  
>> +#ifdef CONFIG_DEBUG_FS
>> +	if (unlikely(!d_swiotlb_usage))
>> +		swiotlb_create_debugfs();
>> +#endif
>> +
>>  	mask = dma_get_seg_boundary(hwdev);
>>  
>>  	tbl_dma_addr &= mask;
>> @@ -528,6 +578,9 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
>>  		dev_warn(hwdev, "swiotlb buffer is full (sz: %zd bytes)\n", size);
>>  	return SWIOTLB_MAP_ERROR;
>>  found:
>> +#ifdef CONFIG_DEBUG_FS
>> +	io_tlb_used += nslots;
>> +#endif
>>  	spin_unlock_irqrestore(&io_tlb_lock, flags);
>>  
>>  	/*
>> @@ -588,6 +641,10 @@ void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr,
>>  		 */
>>  		for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE -1) && io_tlb_list[i]; i--)
>>  			io_tlb_list[i] = ++count;
>> +
>> +#ifdef CONFIG_DEBUG_FS
>> +		io_tlb_used -= nslots;
>> +#endif
>>  	}
>>  	spin_unlock_irqrestore(&io_tlb_lock, flags);
>>  }
>>
> 
> 

  reply	other threads:[~2018-12-07  5:47 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-12-06  3:59 [PATCH RFC 1/1] swiotlb: add debugfs to track swiotlb buffer usage Dongli Zhang
2018-12-06 16:12 ` Joe Jin
2018-12-07  5:49   ` Dongli Zhang [this message]
2018-12-07  6:33     ` Joe Jin
2018-12-07 13:17     ` Robin Murphy
2018-12-09  1:37       ` Dongli Zhang

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=28a9f44a-d1bf-fb34-5a57-cfdb7bb23163@oracle.com \
    --to=dongli.zhang@oracle.com \
    --cc=hch@lst.de \
    --cc=iommu@lists.linux-foundation.org \
    --cc=joe.jin@oracle.com \
    --cc=konrad.wilk@oracle.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=m.szyprowski@samsung.com \
    --cc=robin.murphy@arm.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).