[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1321640117.15493.63.camel@shinybook.infradead.org>
Date: Fri, 18 Nov 2011 18:15:17 +0000
From: David Woodhouse <dwmw2@...radead.org>
To: Alex Williamson <alex.williamson@...hat.com>
Cc: rajesh.sankaran@...el.com, iommu@...ts.linux-foundation.org,
linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
chrisw@...s-sol.org, ddutile@...hat.com
Subject: Re: [PATCH] intel-iommu: Manage iommu_coherency globally
On Fri, 2011-11-18 at 10:15 -0700, Alex Williamson wrote:
> A bit heavy handed, but obviously easier. It feels like we could safely
> be more strategic, but maybe we'd end up trashing the cache anyway in a
> drawn out attempt to flush the context and all page tables. However, do
> we actually need a wbinvd_on_all_cpus()? Probably better to trash one
> cache than all of them. Thanks,
Yes, it would be wbinvd_on_all_cpus(). The page-table-walk code would
look something like this...
static void flush_domain_ptes(struct dmar_domain *domain)
{
int level = agaw_to_level(domain->agaw);
struct dma_pte *ptelvl[level + 2], *pte;
pte = domain->pgd;
clflush_cache_range(pte, VTD_PAGE_SIZE);
/* Keep the end condition nice and simple... */
ptelvl[level + 1] = NULL;
while (pte) {
/* BUG_ON(level < 2); */
if (dma_pte_present(pte)) {
if (level == 2) {
/* Flush the bottom-level page, but no need to
process its *contents* any further. */
clflush_cache_range(phys_to_virt(dma_pte_addr(pte)),
VTD_PAGE_SIZE);
} else if (!(pte->val & DMA_PTE_LARGE_PAGE)) {
/* Remember the next PTE at this level; we'll
come back to it. */
ptelvl[level] = pte + 1;
/* Go down a level and process PTEs there. */
pte = phys_to_virt(dma_pte_addr(pte));
clflush_cache_range(pte, VTD_PAGE_SIZE);
level--;
continue;
}
}
pte++;
/* When we reach the end of a PTE page, go back up to the next
level where we came from. */
while (pte && !((unsigned long)pte & ~VTD_PAGE_MASK)) {
level++;
pte = ptelvl[level];
}
}
}
--
dwmw2
Download attachment "smime.p7s" of type "application/x-pkcs7-signature" (5818 bytes)
Powered by blists - more mailing lists