From: Rusty Russell Subject: Re: [RFC PATCH 16/33] Add support for Xen to entry.S. Date: Wed, 19 Jul 2006 05:17:00 +1000 Message-ID: <1153250220.5467.38.camel@localhost.localdomain> References: <20060718091807.467468000@sous-sol.org> <20060718091952.505770000@sous-sol.org> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20060718091952.505770000@sous-sol.org> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Chris Wright Cc: Andrew Morton , Zachary Amsden , Jeremy Fitzhardinge , xen-devel@lists.xensource.com, Jeremy Fitzhardinge , Ian Pratt , linux-kernel@vger.kernel.org, Andi Kleen , virtualization@lists.osdl.org, Christian Limpach List-Id: xen-devel@lists.xenproject.org On Tue, 2006-07-18 at 00:00 -0700, Chris Wright wrote: > plain text document attachment (i386-entry.S) > - change cli/sti > - change test for user mode return to work for kernel mode in ring1 > - check hypervisor saved event mask on return from exception > - add entry points for the hypervisor upcall handlers > - avoid math emulation check when running on Xen > - add nmi handler for running on Xen > > Signed-off-by: Ian Pratt > Signed-off-by: Christian Limpach > Signed-off-by: Jeremy Fitzhardinge > Signed-off-by: Chris Wright > > --- > arch/i386/kernel/asm-offsets.c | 26 +++++++ > arch/i386/kernel/entry.S | 141 ++++++++++++++++++++++++++++++++++++----- > arch/i386/mach-xen/setup-xen.c | 19 +++++ > drivers/xen/core/features.c | 2 > 4 files changed, 169 insertions(+), 19 deletions(-) > > diff -r 5cca1805b8a7 arch/i386/kernel/entry.S > --- a/arch/i386/kernel/entry.S Tue Jul 18 02:20:39 2006 -0400 > +++ b/arch/i386/kernel/entry.S Tue Jul 18 02:22:56 2006 -0400 > @@ -76,8 +76,39 @@ NT_MASK = 0x00004000 > NT_MASK = 0x00004000 > VM_MASK = 0x00020000 > > +#ifndef CONFIG_XEN > +#define DISABLE_INTERRUPTS cli > +#define ENABLE_INTERRUPTS sti > +#else > +#include > + > +EVENT_MASK = 0x2E > + > +/* Offsets into shared_info_t. */ > +#define evtchn_upcall_pending /* 0 */ > +#define evtchn_upcall_mask 1 Erk... Can we get these into asm-offsets? > + > +#ifdef CONFIG_SMP > +/* Set %esi to point to the appropriate vcpu structure */ > +#define GET_VCPU_INFO movl TI_cpu(%ebp),%esi ; \ > + shl $SIZEOF_VCPU_INFO_SHIFT,%esi ; \ > + addl HYPERVISOR_shared_info,%esi > +#else > +#define GET_VCPU_INFO movl HYPERVISOR_shared_info,%esi > +#endif > + > +/* The following end up using/clobbering %esi, because of GET_VCPU_INFO */ > +#define __DISABLE_INTERRUPTS movb $1,evtchn_upcall_mask(%esi) > +#define DISABLE_INTERRUPTS GET_VCPU_INFO ; \ > + __DISABLE_INTERRUPTS > +#define ENABLE_INTERRUPTS GET_VCPU_INFO ; \ > + movb $0,evtchn_upcall_mask(%esi) > +#define __TEST_PENDING testb $0xFF,evtchn_upcall_pending(%esi) > +#endif > + > + Actually, is it possible to move these to a header somewhere? In the paravirt_ops patches I used the names CLI and STI (copied from the VMI patches), but didn't allow them to clobber any regs. They're defined in asm-i386/paravirt.h. If we want to allow them to clobber %esi, then your names are probably better, but we should clobber %esi in the !CONFIG_XEN case too if CONFIG_DEBUG_KERNEL. Rusty. -- Help! Save Australia from the worst of the DMCA: http://linux.org.au/law