[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH v2 1/5] arm/xen: define xen_remap as ioremap_cached



On Tue, 4 Jun 2013, Catalin Marinas wrote:
> On Tue, Jun 04, 2013 at 10:20:50AM +0100, Ian Campbell wrote:
> > On Mon, 2013-06-03 at 16:33 +0100, Stefano Stabellini wrote:
> > > Define xen_remap as ioremap_cache (MT_MEMORY and MT_DEVICE_CACHED end up
> > > having the same AttrIndx encoding).
> > 
> > The entries in static struct mem_type mem_types[] look entirely
> > different to me.  What am I missing?
> >     [MT_DEVICE_CACHED] = {    /* ioremap_cached */
> >             .prot_pte       = PROT_PTE_DEVICE | L_PTE_MT_DEV_CACHED,
> >             .prot_l1        = PMD_TYPE_TABLE,
> >             .prot_sect      = PROT_SECT_DEVICE | PMD_SECT_WB,
> >             .domain         = DOMAIN_IO,
> >     },
> >     [MT_MEMORY] = {
> >             .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY,
> >             .prot_l1   = PMD_TYPE_TABLE,
> >             .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
> >             .domain    = DOMAIN_KERNEL,
> >     },
> > 
> > I can see in pgtable-3level.h how L_PTE_MT_DEV_CACHED and
> > L_PTE_MT_WRITEBACK are the same but not where the MT_WRITEBACK comes
> > from for MT_MEMORY. Things are less clear in pgtable-2level.h, where one
> > is 0x3 and the other is 0xb. I can see that the entries are the same in
> > armv6_mt_table but how that would apply to a v7 processor?
> 
> PROT_PTE_DEVICE and PROT_SECT_DEVICE above don't contain any memory type
> information, just attributes/permission - present, young, dirty and XN:
> 
> #define PROT_PTE_DEVICE               
> L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_XN
> #define PROT_SECT_DEVICE      PMD_TYPE_SECT|PMD_SECT_AP_WRITE
> 
> The memory type is given by the L_PTE_MT_DEV_CACHED and PMD_SECT_WB
> macros. Let's take prot_sect first as it's simpler. For MT_DEVICE_CACHED
> we have:
> 
> .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE | PMD_SECT_WB
> 
> For MT_MEMORY we have:
> 
> .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE
> 
> The cache policy is added later to MT_MEMORY which is either WB or WBWA
> (based on SMP, no particular reason as these are just processor hints;
> for some historical reasons we enabled WBWA for ARM11MPCore but we could
> leave it on all the time).
> 
> Similarly for prot_pte, present, young, dirty are the same.
> 
> Regarding the type, on ARMv7 (with or without LPAE) we use TEX remapping
> and L_PTE_MT_DEVICE has the same index (3-bit TEX[0], C, B for NMRR/PRRR
> or TEX[2:0] for MAIR0/MAIR1 registers) as Normal Cacheable Writeback
> memory (there is no such thing as Device memory with cacheability
> attributes, only Normal Cacheable memory).
> 
> We have XN in addition for MT_DEVICE_CACHED to prevent speculative
> instruction fetches. However, you still get speculative D-cache line
> fills since the memory is Normal Cacheable.
> 
> > Anyhow, even if I'm prepared to believe that MT_MEMORY and
> > MT_DEVICE_CACHED end up being the same thing (which TBH I am) it seems
> > that the level of abstraction involved makes us vulnerable to future
> > changes subtly breaking things for us. What about:
> > 
> >         /* Device shared memory */
> >         #define ioremap_shm(cookie,size)            __arm_ioremap((cookie), 
> > (size), MT_MEMORY)
> 
> For my understanding, what is Xen doing with such mapping? I would
> rather make ioremap_cached() use MT_MEMORY on ARMv6/v7 (or make it
> closer to that, I'm not sure about the implications on ARMv5 and earlier
> but for newer architectures I don't see the point in pretending to have
> Cacheable Device memory). That's however for Russell to comment.

Xen guests share these pages with one another and place a lockless ring
buffer on it for bidirectional communication. So the page that is being
ioremapped actually belongs to another guest and it's RAM.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.