From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konrad Rzeszutek Wilk Subject: [PATCH 4/4] xen/acpi: Prep saved_context cr3 values. Date: Wed, 17 Oct 2012 09:49:46 -0400 Message-ID: <1350481786-4969-5-git-send-email-konrad.wilk@oracle.com> References: <1350481786-4969-1-git-send-email-konrad.wilk@oracle.com> Return-path: In-Reply-To: <1350481786-4969-1-git-send-email-konrad.wilk@oracle.com> Sender: linux-kernel-owner@vger.kernel.org To: linux-kernel@vger.kernel.org, xen-devel@lists.xensource.com, lenb@kernel.org, linux-acpi@vger.kernel.org, hpa@zytor.com, x86@kernel.org Cc: Konrad Rzeszutek Wilk List-Id: linux-acpi@vger.kernel.org When save_processor_state is executed it executes a bunch of pvops calls to save the CPU state values. When it comes back from Xen's S3 (so acpi_enter_sleep_state, which ends up calling xen_acpi_notify_hypervisor_state), it ends up back in the assembler code in wakeup_[32|64].S. It skips the wakeup calls (so wakeup_pmode_return and wakeup_long64) as that has been done in the hypervisor. Instead it continues on in the resume_point (64-bit) or ret_point (32-bit). Most of the calls in there are safe to be executed except when it comes to reloading of cr3 (which it only does on 64-bit kernels). Since they are native assembler calls and Xen expects a PV kernel to prep those to use machine address for cr3 that is what we are going to do. Note: that it is not Machine Frame Numbers (those are used in the MMUEXT_NEW_BASEPTR hypercall for cr3 installation) but the machine physical address. When the assembler code executes this mov %ebx, cr3 it ends end up trapped in the hypervisor (traps.c) which properly now sets the cr3 value. Signed-off-by: Konrad Rzeszutek Wilk --- drivers/xen/acpi.c | 25 +++++++++++++++++++++++++ 1 files changed, 25 insertions(+), 0 deletions(-) diff --git a/drivers/xen/acpi.c b/drivers/xen/acpi.c index 119d42a..25e612c 100644 --- a/drivers/xen/acpi.c +++ b/drivers/xen/acpi.c @@ -35,6 +35,13 @@ #include #include +#include + +#ifdef CONFIG_X86_64 +#include +extern struct saved_context saved_context; +#endif + int xen_acpi_notify_hypervisor_state(u8 sleep_state, u32 pm1a_cnt, u32 pm1b_cnt) { @@ -56,7 +63,25 @@ int xen_acpi_notify_hypervisor_state(u8 sleep_state, pm1a_cnt, pm1b_cnt); return -1; } +#ifdef CONFIG_X86_64 + /* We do not need to anything for 32-bit kernels as the + * low-level calls (write to cr3) is done in the wakup code + * which we never execute when running under Xen. + */ + { + unsigned long mfn; + /* resume_point in wakeup_64.s barrels through and does: + * movq saved_context_cr3(%rax), %rbx + * movq %rbx, %cr3 + * so lets prep the values so they are OK with the + * hypervisor. */ + mfn = pfn_to_mfn(PFN_DOWN(saved_context.cr3)); + /* and back to a physical address */ + mfn = PFN_PHYS(mfn); + saved_context.cr3 = mfn; + } +#endif HYPERVISOR_dom0_op(&op); return 1; } -- 1.7.7.6