tech-kern archive

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

Re: [PAE support] Types + cosmetic fixes



On 02/25/10 21:37, Jason Thorpe wrote:
On Feb 24, 2010, at 4:51 PM, Jean-Yves Migeon wrote:
On 02/25/10 00:42, Jason Thorpe wrote:
On Feb 23, 2010, at 7:57 AM, Jean-Yves Migeon wrote:
Yep, i386-pae. IMHO, modules cannot be "safely" shared between PAE and
non-PAE.

Why not?

bus_dma/bus_space(9), like drm(4). Using non-PAE modules that manipulate 32 
bits bus_addr_t with a kernel expecting 64 bits ones (PAE), mixed with some DMA 
in: I would expect nasty, unexpected breakage.

The right solution is to find a unified ABI that works for both PAE and 
non-PAE.  The i386 port's bus_addr_t should probably just be made 64-bit.

Making paddr_t 64 bits for both PAE and non-PAE is a start.

However, I won't make it in such a short timeframe, as it envisions a broader view (ABI compatibility between i386/i386-pae/xen, for modules). I am first merging the PAE patches and fixes from jmorse and me then my Xen work; some parts being redundant between those two.

PAE would remain as an option for GENERIC, and need a kernel recompile (at least, for 5.99.x) + build.sh modules.

IMHO, the discussion encompasses how kernel and modules inter-dependancy should be expressed (in terms of API/ABI, kernel revision, etc.) Way over my technical skill at the present time, unfortunately; this requires much thought :o

--
Jean-Yves Migeon
jeanyves.migeon%free.fr@localhost




Home | Main Index | Thread Index | Old Index