From: David Woodhouse <dwmw2@infradead.org>
To: Fenghua Yu <fenghua.yu@intel.com>
Cc: Tony Luck <tony.luck@intel.com>,
iommu@lists.linux-foundation.org, linux-ia64@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/4] Bug Fix drivers/pci/intel-iommu.c: correct sglist
Date: Wed, 05 Aug 2009 07:59:10 +0000 [thread overview]
Message-ID: <1249459150.9324.346.camel@macbook.infradead.org> (raw)
In-Reply-To: <20090804220937.GA17945@linux-os.sc.intel.com>
On Tue, 2009-08-04 at 15:09 -0700, Fenghua Yu wrote:
>
> - for_each_sg(sglist, sg, nelems, i)
> - size += aligned_nrpages(sg->offset, sg->length);
> + for_each_sg(sglist, sg, nelems, i) {
> + dma_addr_t addr = sg_phys(sg);
> + size += aligned_nrpages(addr, sg->length);
> + }
What's the point in this part? We mask off the high bits of the first
argument to aligned_nrpages() -- we're only interested in the offset
within the (mm) page. So sg->offset is fine; we don't need to use
sg_phys(), surely? And this is in code where we're desperately trying to
shave off as many cycles as we can...
This version of the patch should suffice, yes?
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index ebc9b8d..ed6109d 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -1648,6 +1648,15 @@ static int domain_context_mapped(struct pci_dev *pdev)
tmp->devfn);
}
+/* Returns a number of VTD pages, but aligned to MM page size */
+static inline unsigned long aligned_nrpages(unsigned long host_addr,
+ size_t size)
+{
+ host_addr &= ~PAGE_MASK;
+ return PAGE_ALIGN(host_addr + size) >> VTD_PAGE_SHIFT;
+}
+
+
static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
struct scatterlist *sg, unsigned long phys_pfn,
unsigned long nr_pages, int prot)
@@ -1675,7 +1684,7 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
uint64_t tmp;
if (!sg_res) {
- sg_res = (sg->offset + sg->length + VTD_PAGE_SIZE - 1) >> VTD_PAGE_SHIFT;
+ sg_res = aligned_nrpages(sg->offset, sg->length);
sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset;
sg->dma_length = sg->length;
pteval = page_to_phys(sg_page(sg)) | prot;
@@ -2415,14 +2424,6 @@ error:
return ret;
}
-/* Returns a number of VTD pages, but aligned to MM page size */
-static inline unsigned long aligned_nrpages(unsigned long host_addr,
- size_t size)
-{
- host_addr &= ~PAGE_MASK;
- return PAGE_ALIGN(host_addr + size) >> VTD_PAGE_SHIFT;
-}
-
/* This takes a number of _MM_ pages, not VTD pages */
static struct iova *intel_alloc_iova(struct device *dev,
struct dmar_domain *domain,
@@ -2875,7 +2876,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int ne
start_vpfn = mm_to_dma_pfn(iova->pfn_lo);
- ret = domain_sg_mapping(domain, start_vpfn, sglist, mm_to_dma_pfn(size), prot);
+ ret = domain_sg_mapping(domain, start_vpfn, sglist, size, prot);
if (unlikely(ret)) {
/* clear the page */
dma_pte_clear_range(domain, start_vpfn,
--
dwmw2
WARNING: multiple messages have this Message-ID (diff)
From: David Woodhouse <dwmw2@infradead.org>
To: Fenghua Yu <fenghua.yu@intel.com>
Cc: Tony Luck <tony.luck@intel.com>,
iommu@lists.linux-foundation.org, linux-ia64@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/4] Bug Fix drivers/pci/intel-iommu.c: correct sglist size calculation
Date: Wed, 05 Aug 2009 08:59:10 +0100 [thread overview]
Message-ID: <1249459150.9324.346.camel@macbook.infradead.org> (raw)
In-Reply-To: <20090804220937.GA17945@linux-os.sc.intel.com>
On Tue, 2009-08-04 at 15:09 -0700, Fenghua Yu wrote:
>
> - for_each_sg(sglist, sg, nelems, i)
> - size += aligned_nrpages(sg->offset, sg->length);
> + for_each_sg(sglist, sg, nelems, i) {
> + dma_addr_t addr = sg_phys(sg);
> + size += aligned_nrpages(addr, sg->length);
> + }
What's the point in this part? We mask off the high bits of the first
argument to aligned_nrpages() -- we're only interested in the offset
within the (mm) page. So sg->offset is fine; we don't need to use
sg_phys(), surely? And this is in code where we're desperately trying to
shave off as many cycles as we can...
This version of the patch should suffice, yes?
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index ebc9b8d..ed6109d 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -1648,6 +1648,15 @@ static int domain_context_mapped(struct pci_dev *pdev)
tmp->devfn);
}
+/* Returns a number of VTD pages, but aligned to MM page size */
+static inline unsigned long aligned_nrpages(unsigned long host_addr,
+ size_t size)
+{
+ host_addr &= ~PAGE_MASK;
+ return PAGE_ALIGN(host_addr + size) >> VTD_PAGE_SHIFT;
+}
+
+
static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
struct scatterlist *sg, unsigned long phys_pfn,
unsigned long nr_pages, int prot)
@@ -1675,7 +1684,7 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
uint64_t tmp;
if (!sg_res) {
- sg_res = (sg->offset + sg->length + VTD_PAGE_SIZE - 1) >> VTD_PAGE_SHIFT;
+ sg_res = aligned_nrpages(sg->offset, sg->length);
sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset;
sg->dma_length = sg->length;
pteval = page_to_phys(sg_page(sg)) | prot;
@@ -2415,14 +2424,6 @@ error:
return ret;
}
-/* Returns a number of VTD pages, but aligned to MM page size */
-static inline unsigned long aligned_nrpages(unsigned long host_addr,
- size_t size)
-{
- host_addr &= ~PAGE_MASK;
- return PAGE_ALIGN(host_addr + size) >> VTD_PAGE_SHIFT;
-}
-
/* This takes a number of _MM_ pages, not VTD pages */
static struct iova *intel_alloc_iova(struct device *dev,
struct dmar_domain *domain,
@@ -2875,7 +2876,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int ne
start_vpfn = mm_to_dma_pfn(iova->pfn_lo);
- ret = domain_sg_mapping(domain, start_vpfn, sglist, mm_to_dma_pfn(size), prot);
+ ret = domain_sg_mapping(domain, start_vpfn, sglist, size, prot);
if (unlikely(ret)) {
/* clear the page */
dma_pte_clear_range(domain, start_vpfn,
--
dwmw2
next prev parent reply other threads:[~2009-08-05 7:59 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-08-02 8:45 [PATCH 3/15] arch/ia64: Use DIV_ROUND_CLOSEST Julia Lawall
2009-08-02 8:45 ` Julia Lawall
2009-08-04 22:09 ` [PATCH 1/4] Bug Fix drivers/pci/intel-iommu.c: correct sglist size calculation Fenghua Yu
2009-08-04 22:09 ` Fenghua Yu
2009-08-05 7:59 ` David Woodhouse [this message]
2009-08-05 7:59 ` David Woodhouse
2009-08-05 23:22 ` [PATCH 3/15] arch/ia64: Use DIV_ROUND_CLOSEST Yu, Fenghua
2009-08-05 23:22 ` Yu, Fenghua
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=1249459150.9324.346.camel@macbook.infradead.org \
--to=dwmw2@infradead.org \
--cc=fenghua.yu@intel.com \
--cc=iommu@lists.linux-foundation.org \
--cc=linux-ia64@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=tony.luck@intel.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.