Re: [PATCH v5] modify the IO_TLB_SEGSIZE and IO_TLB_DEFAULT_SIZE configurable as flexible requirement about SW-IOMMU.

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Tue, Mar 03, 2015 at 04:11:09PM +0800, Wang Xiaoming wrote:
> The maximum of SW-IOMMU is limited to 2^11*128 = 256K.
> And the size of IO_TLB_DEFAULT_SIZE is limited to (64UL<<20) 64M now.
> While in different platform and different requirement this seems improper.
> So modifing the IO_TLB_SEGSIZE to io_tlb_segsize and IO_TLB_DEFAULT_SIZE
> to io_tlb_default_size which can configure by kernel cmdline.
> This can meet different requirement.
> 
> Signed-off-by: Chuansheng Liu <chuansheng.liu@intel.com>
> Signed-off-by: Zhang Dongxing <dongxing.zhang@intel.com>
> Signed-off-by: Wang Xiaoming <xiaoming.wang@intel.com>
> ---
> patch v1 make this change at Kconfig
> which needs to edit the .config manually.
> https://lkml.org/lkml/2015/1/25/571
> 
> patch v2 only change IO_TLB_SEGSIZE configurable.
> https://lkml.org/lkml/2015/2/5/812
> 
> patch v3 parsing io_tlb_segsize and 
> io_tlb_default_size independently.
> https://lkml.org/lkml/2015/2/15/217
> 
> patch v4 hasn't validated the data from
> command line.

Thank you for redoing this per review.

> https://lkml.org/lkml/2015/2/17/114
> 
>  Documentation/kernel-parameters.txt  |    9 ++++-
>  arch/mips/cavium-octeon/dma-octeon.c |    2 +-
>  arch/mips/netlogic/common/nlm-dma.c  |    2 +-
>  drivers/xen/swiotlb-xen.c            |    6 +--
>  include/linux/swiotlb.h              |    8 +---
>  lib/swiotlb.c                        |   68 +++++++++++++++++++++++++---------
>  6 files changed, 65 insertions(+), 30 deletions(-)
> 
> diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
> index 4df73da..1f50e86 100644
> --- a/Documentation/kernel-parameters.txt
> +++ b/Documentation/kernel-parameters.txt
> @@ -3438,10 +3438,17 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
>  			it if 0 is given (See Documentation/cgroups/memory.txt)
>  
>  	swiotlb=	[ARM,IA-64,PPC,MIPS,X86]
> -			Format: { <int> | force }
> +			Format: { <int> | force | <int> | <int>}
                                                        ,

s/|/,/

>  			<int> -- Number of I/O TLB slabs
>  			force -- force using of bounce buffers even if they
>  			         wouldn't be automatically used by the kernel
> +			<int> -- Maximum allowable number of contiguous slabs to map
> +			<int> -- The size of SW-MMU mapped.
> +			Using "," to separate them one by one.

"Use ',' to seperate them."

> +			Example:
> +			BOARD_KERNEL_CMDLINE += swiotlb=32768,force,512,268435456
> +			io_tlb_nslabs=32768, swiotlb_force=1, 
> +			io_tlb_segsize=512, io_tlb_default_size=268435456

I think you can remove the example - and just have it in the C code.

>  
>  	switches=	[HW,M68k]
>  
> diff --git a/arch/mips/cavium-octeon/dma-octeon.c b/arch/mips/cavium-octeon/dma-octeon.c
> index 3778655..a521af6 100644
> --- a/arch/mips/cavium-octeon/dma-octeon.c
> +++ b/arch/mips/cavium-octeon/dma-octeon.c
> @@ -312,7 +312,7 @@ void __init plat_swiotlb_setup(void)
>  		swiotlbsize = 64 * (1<<20);
>  #endif
>  	swiotlb_nslabs = swiotlbsize >> IO_TLB_SHIFT;
> -	swiotlb_nslabs = ALIGN(swiotlb_nslabs, IO_TLB_SEGSIZE);
> +	swiotlb_nslabs = ALIGN(swiotlb_nslabs, io_tlb_segsize);
>  	swiotlbsize = swiotlb_nslabs << IO_TLB_SHIFT;
>  
>  	octeon_swiotlb = alloc_bootmem_low_pages(swiotlbsize);
> diff --git a/arch/mips/netlogic/common/nlm-dma.c b/arch/mips/netlogic/common/nlm-dma.c
> index f3d4ae8..eeffa8f 100644
> --- a/arch/mips/netlogic/common/nlm-dma.c
> +++ b/arch/mips/netlogic/common/nlm-dma.c
> @@ -99,7 +99,7 @@ void __init plat_swiotlb_setup(void)
>  
>  	swiotlbsize = 1 << 20; /* 1 MB for now */
>  	swiotlb_nslabs = swiotlbsize >> IO_TLB_SHIFT;
> -	swiotlb_nslabs = ALIGN(swiotlb_nslabs, IO_TLB_SEGSIZE);
> +	swiotlb_nslabs = ALIGN(swiotlb_nslabs, io_tlb_segsize);
>  	swiotlbsize = swiotlb_nslabs << IO_TLB_SHIFT;
>  
>  	nlm_swiotlb = alloc_bootmem_low_pages(swiotlbsize);
> diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
> index 810ad41..3b3e9fe 100644
> --- a/drivers/xen/swiotlb-xen.c
> +++ b/drivers/xen/swiotlb-xen.c
> @@ -164,11 +164,11 @@ xen_swiotlb_fixup(void *buf, size_t size, unsigned long nslabs)
>  	dma_addr_t dma_handle;
>  	phys_addr_t p = virt_to_phys(buf);
>  
> -	dma_bits = get_order(IO_TLB_SEGSIZE << IO_TLB_SHIFT) + PAGE_SHIFT;
> +	dma_bits = get_order(io_tlb_segsize << IO_TLB_SHIFT) + PAGE_SHIFT;
>  
>  	i = 0;
>  	do {
> -		int slabs = min(nslabs - i, (unsigned long)IO_TLB_SEGSIZE);
> +		int slabs = min(nslabs - i, (unsigned long)io_tlb_segsize);
>  
>  		do {
>  			rc = xen_create_contiguous_region(
> @@ -187,7 +187,7 @@ static unsigned long xen_set_nslabs(unsigned long nr_tbl)
>  {
>  	if (!nr_tbl) {
>  		xen_io_tlb_nslabs = (64 * 1024 * 1024 >> IO_TLB_SHIFT);
> -		xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, IO_TLB_SEGSIZE);
> +		xen_io_tlb_nslabs = ALIGN(xen_io_tlb_nslabs, io_tlb_segsize);
>  	} else
>  		xen_io_tlb_nslabs = nr_tbl;
>  
> diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
> index e7a018e..13506db 100644
> --- a/include/linux/swiotlb.h
> +++ b/include/linux/swiotlb.h
> @@ -8,13 +8,7 @@ struct dma_attrs;
>  struct scatterlist;
>  
>  extern int swiotlb_force;
> -
> -/*
> - * Maximum allowable number of contiguous slabs to map,
> - * must be a power of 2.  What is the appropriate value ?
> - * The complexity of {map,unmap}_single is linearly dependent on this value.
> - */
> -#define IO_TLB_SEGSIZE	128
> +extern int io_tlb_segsize;
>  
>  /*
>   * log of the size of each IO TLB slab.  The number of slabs is command line
> diff --git a/lib/swiotlb.c b/lib/swiotlb.c
> index 4abda07..3b71afd 100644
> --- a/lib/swiotlb.c
> +++ b/lib/swiotlb.c
> @@ -56,6 +56,24 @@
>  int swiotlb_force;
>  
>  /*
> + * default to 128
> + * Maximum allowable number of contiguous slabs to map,
> + * must be a power of 2.  What is the appropriate value ?
> + * define io_tlb_segsize as a parameter
> + * which can be changed dynamically in config file for special usage.
> + * The complexity of {map,unmap}_single is linearly dependent on this value.
> + */
> +#define IO_TLB_SEGSIZE 128

Add the tab back please.

> +int io_tlb_segsize = IO_TLB_SEGSIZE;
> +
> +/* default to 64MB 
> + * define io_tlb_default_size as a parameter
> + * which can be changed dynamically in config file for special usage.

.. or as a parameter during bootup.

> + */
> +#define IO_TLB_DEFAULT_SIZE (64UL<<20)
> +static unsigned long io_tlb_default_size = IO_TLB_DEFAULT_SIZE;
> +
> +/*
>   * Used to do a quick range check in swiotlb_tbl_unmap_single and
>   * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this
>   * API.
> @@ -101,13 +119,32 @@ setup_io_tlb_npages(char *str)
>  {
>  	if (isdigit(*str)) {
>  		io_tlb_nslabs = simple_strtoul(str, &str, 0);
> -		/* avoid tail segment of size < IO_TLB_SEGSIZE */
> -		io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
>  	}
>  	if (*str == ',')
>  		++str;
> -	if (!strcmp(str, "force"))
> +	if (!strncmp(str, "force", 5)) {
>  		swiotlb_force = 1;
> +		str += 5;
> +	}

So the format is now:

	Format: { <int> | force | <int> | <int>}

which means I can do
	32,22323,force

Or
	force,32

Or	
	32,force

I think you need to make function be inside a loop
to deal with 'force' being at odd locations.

> +	if (*str == ',')
> +		++str;
> +	if (isdigit(*str)) {
> +		int n = 0;
> +		io_tlb_segsize = simple_strtoul(str, &str, 0);
> +		io_tlb_segsize = ALIGN(io_tlb_segsize, IO_TLB_SEGSIZE);
> +		while ((io_tlb_segsize - 1) >> n)
> +			n++;
> +		io_tlb_segsize = (1 << n);
> +	}
> +	if (*str == ',')
> +		++str;
> +	if (isdigit(*str)) {
> +		io_tlb_default_size = simple_strtoul(str, &str, 0);
> +		io_tlb_default_size = ALIGN(io_tlb_default_size, IO_TLB_DEFAULT_SIZE);
> +	}
> +
> +	/* avoid tail segment of size < io_tlb_segsize */
> +	io_tlb_nslabs = ALIGN(io_tlb_nslabs, io_tlb_segsize);
>  
>  	return 0;
>  }
> @@ -120,15 +157,13 @@ unsigned long swiotlb_nr_tbl(void)
>  }
>  EXPORT_SYMBOL_GPL(swiotlb_nr_tbl);
>  
> -/* default to 64MB */
> -#define IO_TLB_DEFAULT_SIZE (64UL<<20)
>  unsigned long swiotlb_size_or_default(void)
>  {
>  	unsigned long size;
>  
>  	size = io_tlb_nslabs << IO_TLB_SHIFT;
>  
> -	return size ? size : (IO_TLB_DEFAULT_SIZE);
> +	return size ? size : (io_tlb_default_size);
>  }
>  
>  /* Note that this doesn't work with highmem page */
> @@ -183,7 +218,7 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
>  
>  	/*
>  	 * Allocate and initialize the free list array.  This array is used
> -	 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
> +	 * to find contiguous free memory regions of size up to io_tlb_segsize
>  	 * between io_tlb_start and io_tlb_end.
>  	 */
>  	io_tlb_list = memblock_virt_alloc(
> @@ -193,7 +228,7 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
>  				PAGE_ALIGN(io_tlb_nslabs * sizeof(phys_addr_t)),
>  				PAGE_SIZE);
>  	for (i = 0; i < io_tlb_nslabs; i++) {
> -		io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE);
> +		io_tlb_list[i] = io_tlb_segsize - OFFSET(i, io_tlb_segsize);
>  		io_tlb_orig_addr[i] = INVALID_PHYS_ADDR;
>  	}
>  	io_tlb_index = 0;
> @@ -211,13 +246,12 @@ int __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
>  void  __init
>  swiotlb_init(int verbose)
>  {
> -	size_t default_size = IO_TLB_DEFAULT_SIZE;
>  	unsigned char *vstart;
>  	unsigned long bytes;
>  
>  	if (!io_tlb_nslabs) {
> -		io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
> -		io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
> +		io_tlb_nslabs = (io_tlb_default_size >> IO_TLB_SHIFT);
> +		io_tlb_nslabs = ALIGN(io_tlb_nslabs, io_tlb_segsize);
>  	}
>  
>  	bytes = io_tlb_nslabs << IO_TLB_SHIFT;
> @@ -249,7 +283,7 @@ swiotlb_late_init_with_default_size(size_t default_size)
>  
>  	if (!io_tlb_nslabs) {
>  		io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
> -		io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
> +		io_tlb_nslabs = ALIGN(io_tlb_nslabs, io_tlb_segsize);
>  	}
>  
>  	/*
> @@ -308,7 +342,7 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs)
>  
>  	/*
>  	 * Allocate and initialize the free list array.  This array is used
> -	 * to find contiguous free memory regions of size up to IO_TLB_SEGSIZE
> +	 * to find contiguous free memory regions of size up to io_tlb_segsize
>  	 * between io_tlb_start and io_tlb_end.
>  	 */
>  	io_tlb_list = (unsigned int *)__get_free_pages(GFP_KERNEL,
> @@ -324,7 +358,7 @@ swiotlb_late_init_with_tbl(char *tlb, unsigned long nslabs)
>  		goto cleanup4;
>  
>  	for (i = 0; i < io_tlb_nslabs; i++) {
> -		io_tlb_list[i] = IO_TLB_SEGSIZE - OFFSET(i, IO_TLB_SEGSIZE);
> +		io_tlb_list[i] = io_tlb_segsize - OFFSET(i, io_tlb_segsize);
>  		io_tlb_orig_addr[i] = INVALID_PHYS_ADDR;
>  	}
>  	io_tlb_index = 0;
> @@ -493,7 +527,7 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
>  
>  			for (i = index; i < (int) (index + nslots); i++)
>  				io_tlb_list[i] = 0;
> -			for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE - 1) && io_tlb_list[i]; i--)
> +			for (i = index - 1; (OFFSET(i, io_tlb_segsize) != io_tlb_segsize - 1) && io_tlb_list[i]; i--)
>  				io_tlb_list[i] = ++count;
>  			tlb_addr = io_tlb_start + (index << IO_TLB_SHIFT);
>  
> @@ -571,7 +605,7 @@ void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr,
>  	 */
>  	spin_lock_irqsave(&io_tlb_lock, flags);
>  	{
> -		count = ((index + nslots) < ALIGN(index + 1, IO_TLB_SEGSIZE) ?
> +		count = ((index + nslots) < ALIGN(index + 1, io_tlb_segsize) ?
>  			 io_tlb_list[index + nslots] : 0);
>  		/*
>  		 * Step 1: return the slots to the free list, merging the
> @@ -585,7 +619,7 @@ void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr,
>  		 * Step 2: merge the returned slots with the preceding slots,
>  		 * if available (non zero)
>  		 */
> -		for (i = index - 1; (OFFSET(i, IO_TLB_SEGSIZE) != IO_TLB_SEGSIZE -1) && io_tlb_list[i]; i--)
> +		for (i = index - 1; (OFFSET(i, io_tlb_segsize) != io_tlb_segsize -1) && io_tlb_list[i]; i--)
>  			io_tlb_list[i] = ++count;
>  	}
>  	spin_unlock_irqrestore(&io_tlb_lock, flags);
> -- 
> 1.7.9.5
>