lkml.org 
[lkml]   [2012]   [Jul]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRE: [PATCHv4 2/2] ARM: dma-mapping: remove custom consistent dma region
Date
Hello,

On Monday, July 02, 2012 1:06 PM Hiroshi Doyu wrote:

> On Mon, 25 Jun 2012 10:47:27 +0200
> Marek Szyprowski <m.szyprowski@samsung.com> wrote:
>
> > This patch changes dma-mapping subsystem to use generic vmalloc areas
> > for all consistent dma allocations. This increases the total size limit
> > of the consistent allocations and removes platform hacks and a lot of
> > duplicated code.
> >
> > Atomic allocations are served from special pool preallocated on boot,
> > becasue vmalloc areas cannot be reliably created in atomic context.
> >
> > Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> > Reviewed-by: Kyungmin Park <kyungmin.park@samsung.com>
> > ---
> > Documentation/kernel-parameters.txt | 2 +-
> > arch/arm/include/asm/dma-mapping.h | 2 +-
> > arch/arm/mm/dma-mapping.c | 505 +++++++++++++----------------------
> > arch/arm/mm/mm.h | 3 +
> > include/linux/vmalloc.h | 1 +
> > mm/vmalloc.c | 10 +-
> > 6 files changed, 194 insertions(+), 329 deletions(-)
> >
> ......
> > -static int __init consistent_init(void)
> > -{
> > - int ret = 0;
> > - pgd_t *pgd;
> > - pud_t *pud;
> > - pmd_t *pmd;
> > - pte_t *pte;
> > - int i = 0;
> > - unsigned long base = consistent_base;
> > - unsigned long num_ptes = (CONSISTENT_END - base) >> PMD_SHIFT;
> > -
> > - if (IS_ENABLED(CONFIG_CMA) && !IS_ENABLED(CONFIG_ARM_DMA_USE_IOMMU))
> > - return 0;
> > -
> > - consistent_pte = kmalloc(num_ptes * sizeof(pte_t), GFP_KERNEL);
> > - if (!consistent_pte) {
> > - pr_err("%s: no memory\n", __func__);
> > - return -ENOMEM;
> > - }
> > -
> > - pr_debug("DMA memory: 0x%08lx - 0x%08lx:\n", base, CONSISTENT_END);
> > - consistent_head.vm_start = base;
> > -
> > - do {
> > - pgd = pgd_offset(&init_mm, base);
> > -
> > - pud = pud_alloc(&init_mm, pgd, base);
> > - if (!pud) {
> > - pr_err("%s: no pud tables\n", __func__);
> > - ret = -ENOMEM;
> > - break;
> > - }
> > -
> > - pmd = pmd_alloc(&init_mm, pud, base);
> > - if (!pmd) {
> > - pr_err("%s: no pmd tables\n", __func__);
> > - ret = -ENOMEM;
> > - break;
> > - }
> > - WARN_ON(!pmd_none(*pmd));
> > -
> > - pte = pte_alloc_kernel(pmd, base);
> > - if (!pte) {
> > - pr_err("%s: no pte tables\n", __func__);
> > - ret = -ENOMEM;
> > - break;
> > - }
> > -
> > - consistent_pte[i++] = pte;
> > - base += PMD_SIZE;
> > - } while (base < CONSISTENT_END);
> > -
> > - return ret;
> > -}
> > -core_initcall(consistent_init);
> > -
> > static void *__alloc_from_contiguous(struct device *dev, size_t size,
> > pgprot_t prot, struct page **ret_page);
> >
> > -static struct arm_vmregion_head coherent_head = {
> > - .vm_lock = __SPIN_LOCK_UNLOCKED(&coherent_head.vm_lock),
> > - .vm_list = LIST_HEAD_INIT(coherent_head.vm_list),
> > +static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp,
> > + pgprot_t prot, struct page **ret_page,
> > + const void *caller);
> > +
> > +static void *
> > +__dma_alloc_remap(struct page *page, size_t size, gfp_t gfp, pgprot_t prot,
> > + const void *caller)
> > +{
> > + struct vm_struct *area;
> > + unsigned long addr;
> > +
> > + area = get_vm_area_caller(size, VM_ARM_DMA_CONSISTENT | VM_USERMAP,
> > + caller);
> > + if (!area)
> > + return NULL;
> > + addr = (unsigned long)area->addr;
> > + area->phys_addr = __pfn_to_phys(page_to_pfn(page));
> > +
> > + if (ioremap_page_range(addr, addr + size, area->phys_addr, prot)) {
> > + vunmap((void *)addr);
> > + return NULL;
> > + }
> > + return (void *)addr;
> > +}
>
> The above "ioremap_page_range()" seems to be executed against normal
> pages(liner kernel mapping) with setting a new prot, because pages were
> passed from __dma_alloc_buffer(){..alloc_pages()...}. For me, this is
> creating another page mapping with different pgprot, and it can cause
> the pgprot inconsistency. This reminds me of the following old patch.
>
> [RFC PATCH] Avoid aliasing mappings in DMA coherent allocator
> http://lists.infradead.org/pipermail/linux-arm-kernel/2012-June/106815.html

If I remember correctly that approach has been dropped because:
a) it consumed a fixed, quite large amount of RAM only for DMA mapping purposes
what was considered as a waste of resources
b) didn't work with some hardware configurations which had DMA zone less than 2MiB.

> I think that this is why ioremap() isn't allowed with RAM.
>
> __arm_ioremap_pfn_caller() doens't allow RAM remapping.
>
> 193 void __iomem * __arm_ioremap_pfn_caller(unsigned long pfn,
> 194 unsigned long offset, size_t size, unsigned int mtype, void *caller)
> 195 {
> 196 const struct mem_type *type;
> 197 int err;
> ... .
> 240 /*
> 241 * Don't allow RAM to be mapped - this causes problems with ARMv6+
> 242 */
> 243 if (WARN_ON(pfn_valid(pfn)))
> 244 return NULL;
> ...
>
> So my question is:
> 1, is the above ioremap_page_range() creating another page mapping
> with a new pgprot, in addition to liner mapping?

Yes it does. My patch does exactly the same thing a the existing __dma_alloc_remap()
by using a generic vmalloc helper functions.

> 2, If so, is it safe for pgprot inconsistency from different vaddrs?

It probably depends on the hardware. Right now, although specification says this is
a violation, no side effects has been observed and such solution is already used for
years.

> I hope that my questins are making sense.

Best regards
--
Marek Szyprowski
Samsung Poland R&D Center




\
 
 \ /
  Last update: 2012-07-02 17:22    [W:0.083 / U:0.248 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site