Re: [Xen-devel] Re: [Xen-changelog] [linux-2.6.18-xen] blktap2: a completely rewritten blktap implementation

[Isaku Yamahata <yamahata@xxxxxxxxxxxxx>]

How about this patch?

blktap2: add tlb flush properly.

xen_invlpg() flushes tlb on its cpu, but tlb flush is needed on
all cpus. So replace xen_invlpg() with more proper ones.
Maybe it would be possible to make tlb flush less.
this patch also makes blktap2 compile on ia64 because xen_invlpg()
is x86 specific.

Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>

diff --git a/drivers/xen/blktap2/device.c b/drivers/xen/blktap2/device.c
--- a/drivers/xen/blktap2/device.c
+++ b/drivers/xen/blktap2/device.c
@@ -3,6 +3,7 @@
 #include <linux/cdrom.h>
 #include <linux/hdreg.h>
 #include <linux/module.h>
+#include <asm/tlbflush.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_ioctl.h>
@@ -163,7 +164,6 @@ blktap_map_uaddr_fn(pte_t *ptep, struct 
 
        BTDBG("ptep %p -> %012llx\n", ptep, pte_val(*pte));
        set_pte(ptep, *pte);
-       xen_invlpg(addr);
        return 0;
 }
 
@@ -182,7 +182,6 @@ blktap_umap_uaddr_fn(pte_t *ptep, struct
 
        BTDBG("ptep %p\n", ptep);
        pte_clear(mm, addr, ptep);
-       xen_invlpg(addr);
        return 0;
 }
 
@@ -322,6 +321,7 @@ blktap_unmap(struct blktap *tap, struct 
                if (request->handles[i].kernel == INVALID_GRANT_HANDLE) {
                        kvaddr = request_to_kaddr(request, i);
                        blktap_umap_uaddr(&init_mm, kvaddr);
+                       flush_tlb_kernel_range(kvaddr, kvaddr + PAGE_SIZE);
                        set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
                                            INVALID_P2M_ENTRY);
                }
@@ -551,7 +551,9 @@ blktap_map(struct blktap *tap,
 
        pte = mk_pte(page, ring->vma->vm_page_prot);
        blktap_map_uaddr(ring->vma->vm_mm, uvaddr, pte_mkwrite(pte));
+       flush_tlb_mm(ring->vma->vm_mm);
        blktap_map_uaddr(&init_mm, kvaddr, mk_pte(page, PAGE_KERNEL));
+       flush_tlb_kernel_range(kvaddr, kvaddr + PAGE_SIZE);
 
        set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT, pte_mfn(pte));
        request->handles[seg].kernel = INVALID_GRANT_HANDLE;


On Wed, May 27, 2009 at 07:18:44PM +0900, Isaku Yamahata wrote:
> On Tue, May 26, 2009 at 11:35:05AM -0700, Xen patchbot-linux-2.6.18-xen wrote:
> > diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/device.c
> > --- /dev/null       Thu Jan 01 00:00:00 1970 +0000
> > +++ b/drivers/xen/blktap2/device.c  Tue May 26 11:23:16 2009 +0100
> > +static int
> > +blktap_map_uaddr_fn(pte_t *ptep, struct page *pmd_page,
> > +               unsigned long addr, void *data)
> > +{
> > +   pte_t *pte = (pte_t *)data;
> > +
> > +   BTDBG("ptep %p -> %012llx\n", ptep, pte_val(*pte));
> > +   set_pte(ptep, *pte);
> > +   xen_invlpg(addr);
> > +   return 0;
> > +}
> ...
> > +static int
> > +blktap_umap_uaddr_fn(pte_t *ptep, struct page *pmd_page,
> > +                unsigned long addr, void *data)
> > +{
> > +   struct mm_struct *mm = (struct mm_struct *)data;
> > +
> > +   BTDBG("ptep %p\n", ptep);
> > +   pte_clear(mm, addr, ptep);
> > +   xen_invlpg(addr);
> > +   return 0;
> > +}
> 
> Hi. Although I looked into this for ia64, I think this is generic.
> xen_invlpg() flushes tlb on its cpu and doesn't on other cpus.
> But those functions set ptes for all cpus.
> So should tlb on all cpus be flushed? Maybe by flush_tlb_page()?
> 
> thanks,

-- 
yamahata

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel