KVM: PCIPT: direct mmio pfn check

KVM: PCIPT: direct mmio pfn check

[benami]

In some cases it is not enough to identify mmio memory slots by
pfn_valid.
This patch adds checking PageReserved before assuming that a pfn is RAM.
This patch applies to the main KVM tree.
It was tested with direct mmio using the PCI pass-through tree.

[PATCH] KVM: PCIPT: direct mmio pfn check

[benami]

From: Ben-Ami Yassour <benami@il.ibm.com>

In some cases it is not enough to identify mmio memory slots by
pfn_valid. This patch adds checking the PageReserved as well.

Signed-off-by: Ben-Ami Yassour <benami@il.ibm.com>
Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
---
 virt/kvm/kvm_main.c |   22 +++++++++++++++-------
 1 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index f9427e2..27b2eff 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -76,6 +76,14 @@ static inline int valid_vcpu(int n)
 	return likely(n >= 0 && n < KVM_MAX_VCPUS);
 }
 
+static inline int is_mmio_pfn(pfn_t pfn)
+{
+	if (pfn_valid(pfn))
+		return PageReserved(pfn_to_page(pfn));
+
+	return true;
+}
+
 /*
  * Switches to specified vcpu, until a matching vcpu_put()
  */
@@ -582,7 +590,7 @@ pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 		}
 
 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
-		BUG_ON(pfn_valid(pfn));
+		BUG_ON(!is_mmio_pfn(pfn));
 	} else
 		pfn = page_to_pfn(page[0]);
 
@@ -596,10 +604,10 @@ struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 	pfn_t pfn;
 
 	pfn = gfn_to_pfn(kvm, gfn);
-	if (pfn_valid(pfn))
+	if (!is_mmio_pfn(pfn))
 		return pfn_to_page(pfn);
 
-	WARN_ON(!pfn_valid(pfn));
+	WARN_ON(is_mmio_pfn(pfn));
 
 	get_page(bad_page);
 	return bad_page;
@@ -615,7 +623,7 @@ EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 
 void kvm_release_pfn_clean(pfn_t pfn)
 {
-	if (pfn_valid(pfn))
+	if (!is_mmio_pfn(pfn))
 		put_page(pfn_to_page(pfn));
 }
 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
@@ -641,7 +649,7 @@ EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 
 void kvm_set_pfn_dirty(pfn_t pfn)
 {
-	if (pfn_valid(pfn)) {
+	if (!is_mmio_pfn(pfn)) {
 		struct page *page = pfn_to_page(pfn);
 		if (!PageReserved(page))
 			SetPageDirty(page);
@@ -651,14 +659,14 @@ EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 
 void kvm_set_pfn_accessed(pfn_t pfn)
 {
-	if (pfn_valid(pfn))
+	if (!is_mmio_pfn(pfn))
 		mark_page_accessed(pfn_to_page(pfn));
 }
 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 
 void kvm_get_pfn(pfn_t pfn)
 {
-	if (pfn_valid(pfn))
+	if (!is_mmio_pfn(pfn))
 		get_page(pfn_to_page(pfn));
 }
 EXPORT_SYMBOL_GPL(kvm_get_pfn);
-- 
1.5.5.1

Re: [PATCH] KVM: PCIPT: direct mmio pfn check

[Andrea Arcangeli]

We must apply the Ben-Ami's patch that completes the pfn-mmio support
with the needed checks for PageReserved.

The way the mem_map has to be allocated there will always be non-ram
pages that have a mem_map array backing and to speedup the pfn_valid
check the remaining non-ram check is achieved with PG_reserved bitflag.

Whenever a non-ram page currently used by linux returns true out of
the pfn_valid && PageReserved(pfn_to_page(pfn)) check, it's a kernel
bug and needs fixing. So the below code is quite final, if numa or
non-numa fails to provide this invariant then it's a sparse bug that
should be fixed in the kernel tree, not worked around in kvm.

On top of Ben-Ami's patch, I also need the below incremental patch
applied because with the reserved-ram patch to allow pci-passthrough
w/o VT-d what happens is that qemu has to do the emulated-DMA on those
non-ram pages (reserved-ram isn't used by linux so is also has to be
marked PageReserved). It's ok to export those PageReserved pages
through vm->fault, the put_page will do nothing on those when the
kvm->mmap vma is teardown. So the kvm export of the guest physical
memory to qemu userland for dma, will work as long as the
-reserved-ram patch will ensure that 'struct page' backing exists for
all pages where there's dma (so the virtualized ram). My current
reserved-ram patch ensures this by reserving the ram early in the e820
map so the initial pagetables are allocated above the kernel .text
relocation and then I make the sparse code think the reserved-ram is
actually available (so struct pages are allocated) and finally I've to
reserve those pages in the bootmem allocator so they remain
PageReserved but with 'struct page' backing. So things works fine.

Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>

Index: virt/kvm/kvm_main.c
--- virt/kvm/kvm_main.c.orig	2008-06-25 02:39:51.000000000 +0200
+++ virt/kvm/kvm_main.c	2008-06-25 02:40:35.000000000 +0200
@@ -604,10 +604,9 @@ struct page *gfn_to_page(struct kvm *kvm
 	pfn_t pfn;
 
 	pfn = gfn_to_pfn(kvm, gfn);
-	if (!is_mmio_pfn(pfn))
+	if (pfn_valid(pfn))
 		return pfn_to_page(pfn);
-
-	WARN_ON(is_mmio_pfn(pfn));
+	WARN_ON(1);
 
 	get_page(bad_page);
 	return bad_page;


Thanks.

On Mon, Jun 23, 2008 at 05:52:17PM +0300, benami@il.ibm.com wrote:
> From: Ben-Ami Yassour <benami@il.ibm.com>
> 
> In some cases it is not enough to identify mmio memory slots by
> pfn_valid. This patch adds checking the PageReserved as well.
> 
> Signed-off-by: Ben-Ami Yassour <benami@il.ibm.com>
> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
> ---
>  virt/kvm/kvm_main.c |   22 +++++++++++++++-------
>  1 files changed, 15 insertions(+), 7 deletions(-)
> 
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index f9427e2..27b2eff 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -76,6 +76,14 @@ static inline int valid_vcpu(int n)
>  	return likely(n >= 0 && n < KVM_MAX_VCPUS);
>  }
>  
> +static inline int is_mmio_pfn(pfn_t pfn)
> +{
> +	if (pfn_valid(pfn))
> +		return PageReserved(pfn_to_page(pfn));
> +
> +	return true;
> +}
> +
>  /*
>   * Switches to specified vcpu, until a matching vcpu_put()
>   */
> @@ -582,7 +590,7 @@ pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
>  		}
>  
>  		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
> -		BUG_ON(pfn_valid(pfn));
> +		BUG_ON(!is_mmio_pfn(pfn));
>  	} else
>  		pfn = page_to_pfn(page[0]);
>  
> @@ -596,10 +604,10 @@ struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
>  	pfn_t pfn;
>  
>  	pfn = gfn_to_pfn(kvm, gfn);
> -	if (pfn_valid(pfn))
> +	if (!is_mmio_pfn(pfn))
>  		return pfn_to_page(pfn);
>  
> -	WARN_ON(!pfn_valid(pfn));
> +	WARN_ON(is_mmio_pfn(pfn));
>  
>  	get_page(bad_page);
>  	return bad_page;
> @@ -615,7 +623,7 @@ EXPORT_SYMBOL_GPL(kvm_release_page_clean);
>  
>  void kvm_release_pfn_clean(pfn_t pfn)
>  {
> -	if (pfn_valid(pfn))
> +	if (!is_mmio_pfn(pfn))
>  		put_page(pfn_to_page(pfn));
>  }
>  EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
> @@ -641,7 +649,7 @@ EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
>  
>  void kvm_set_pfn_dirty(pfn_t pfn)
>  {
> -	if (pfn_valid(pfn)) {
> +	if (!is_mmio_pfn(pfn)) {
>  		struct page *page = pfn_to_page(pfn);
>  		if (!PageReserved(page))
>  			SetPageDirty(page);
> @@ -651,14 +659,14 @@ EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
>  
>  void kvm_set_pfn_accessed(pfn_t pfn)
>  {
> -	if (pfn_valid(pfn))
> +	if (!is_mmio_pfn(pfn))
>  		mark_page_accessed(pfn_to_page(pfn));
>  }
>  EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
>  
>  void kvm_get_pfn(pfn_t pfn)
>  {
> -	if (pfn_valid(pfn))
> +	if (!is_mmio_pfn(pfn))
>  		get_page(pfn_to_page(pfn));
>  }
>  EXPORT_SYMBOL_GPL(kvm_get_pfn);
> -- 
> 1.5.5.1
> 
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andrea Arcangeli]

From: Andrea Arcangeli <andrea@qumranet.com>

This has to be applied to the host kernel and for example specifying a
relocation address of 0x20000000 it will allow to start kvm guests
capable of pci-passthrough up to "-m 512" by passing the
"-reserved-ram" parameter in the command line. There's no risk of
errors from the user thanks to the reserved ranges being provided to
the virtualization software through /proc/iomem. Only you shouldn't
run more than one -reserved-ram kvm quest per system at once.

This works by reserving the ram early in the e820 map so the initial
pagetables are allocated above the kernel .text relocation and then I
make the sparse code think the reserved-ram is actually available (so
struct pages are allocated) and finally I've to reserve those pages in
the bootmem allocator immediately after the bootmem allocator has been
initialized, so they remain PageReserved not used by linux, but with
'struct page' backing so they can still be exported to qemu via device
driver vma->fault (as they can still be the target of any emulated
dma, not all devices will passthrough).

The virtualization software must create for the guest an e820 map that
only includes the "reserved RAM" regions but if the guest touches
memory with guest physical address in the "reserved RAM failed" ranges
it should provide that as ram and map it with a non linear
mapping (in practice the only problem is for the first page at address
0 physical which is usually the bios and no sane OS is doing DMA to
it).

vmx ~ # cat /proc/iomem |head -n 20
00000000-00000fff : reserved RAM failed
00001000-0008ffff : reserved RAM
00090000-00091fff : reserved RAM failed
00092000-0009cfff : reserved RAM
0009d000-0009ffff : reserved
000a0000-000ec16f : reserved RAM failed
000ec170-000fffff : reserved
00100000-1fffffff : reserved RAM
20000000-bff9ffff : System RAM
  20000000-20315f65 : Kernel code
  20315f66-204c3767 : Kernel data
  20557000-205c9eff : Kernel bss
bffa0000-bffaffff : ACPI Tables
bffb0000-bffdffff : ACPI Non-volatile Storage
bffe0000-bffedfff : reserved
bfff0000-bfffffff : reserved
d0000000-dfffffff : PCI Bus 0000:02
  d0000000-dfffffff : 0000:02:00.0
e0000000-efffffff : PCI MMCONFIG 0
  e0000000-efffffff : pnp 00:0c

Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
---

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1198,8 +1198,36 @@ config CRASH_DUMP
 	  (CONFIG_RELOCATABLE=y).
 	  For more details see Documentation/kdump/kdump.txt
 
+config RESERVE_PHYSICAL_START
+	bool "Reserve all RAM below PHYSICAL_START (EXPERIMENTAL)"
+	depends on !RELOCATABLE && X86_64
+	help
+	  This makes the kernel use only RAM above __PHYSICAL_START.
+	  All memory below __PHYSICAL_START will be left unused and
+	  marked as "reserved RAM" in /proc/iomem. The few special
+	  pages that can't be relocated at addresses above
+	  __PHYSICAL_START and that can't be guaranteed to be unused
+	  by the running kernel will be marked "reserved RAM failed"
+	  in /proc/iomem. Those may or may be not used by the kernel
+	  (for example SMP trampoline pages would only be used if
+	  CPU hotplug is enabled).
+
+	  The "reserved RAM" can be mapped by virtualization software
+	  with /dev/mem to create a 1:1 mapping between guest physical
+	  (bus) address and host physical (bus) address. This will
+	  allow PCI passthrough with DMA for the guest using the RAM
+	  with the 1:1 mapping. The only detail to take care of is the
+	  RAM marked "reserved RAM failed". The virtualization
+	  software must create for the guest an e820 map that only
+	  includes the "reserved RAM" regions but if the guest touches
+	  memory with guest physical address in the "reserved RAM
+	  failed" ranges (Linux guest will do that even if the RAM
+	  isn't present in the e820 map), it should provide that as
+	  RAM and map it with a non-linear mapping. This should allow
+	  any Linux kernel to run fine and hopefully any other OS too.
+
 config PHYSICAL_START
-	hex "Physical address where the kernel is loaded" if (EMBEDDED || CRASH_DUMP)
+	hex "Physical address where the kernel is loaded" if (EMBEDDED || CRASH_DUMP || RESERVE_PHYSICAL_START)
 	default "0x1000000" if X86_NUMAQ
 	default "0x200000" if X86_64
 	default "0x100000"
diff --git a/arch/x86/kernel/e820_64.c b/arch/x86/kernel/e820_64.c
--- a/arch/x86/kernel/e820_64.c
+++ b/arch/x86/kernel/e820_64.c
@@ -119,7 +119,31 @@ void __init early_res_to_bootmem(unsigne
 		printk(KERN_INFO "  early res: %d [%lx-%lx] %s\n", i,
 			final_start, final_end - 1, r->name);
 		reserve_bootmem_generic(final_start, final_end - final_start);
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		if (r->start < __PHYSICAL_START)
+			add_memory_region(r->start, r->end - r->start,
+					  E820_RESERVED_RAM_FAILED);
+#endif			
 	}
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+	/* solve E820_RESERVED_RAM vs E820_RESERVED_RAM_FAILED conflicts */
+	update_e820();
+
+	/* now reserve E820_RESERVED_RAM */
+	for (i = 0; i < e820.nr_map; i++) {
+		struct e820entry *ei = &e820.map[i];
+
+		if (ei->type != E820_RESERVED_RAM)
+			continue;
+		final_start = max(start, (unsigned long) ei->addr);
+		final_end = min(end, (unsigned long) (ei->addr + ei->size));
+		if (final_start >= final_end)
+			continue;
+		reserve_bootmem_generic(final_start, final_end - final_start);
+		printk(KERN_INFO " bootmem reserved RAM: [%lx-%lx]\n",
+		       final_start, final_end - 1);
+	}
+#endif
 }
 
 /* Check for already reserved areas */
@@ -336,6 +360,16 @@ void __init e820_reserve_resources(void)
 		case E820_RAM:	res->name = "System RAM"; break;
 		case E820_ACPI:	res->name = "ACPI Tables"; break;
 		case E820_NVS:	res->name = "ACPI Non-volatile Storage"; break;
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		case E820_RESERVED_RAM_FAILED:
+			res->name = "reserved RAM failed";
+			break;
+		case E820_RESERVED_RAM:
+			memset(__va(e820.map[i].addr),
+			       POISON_FREE_INITMEM, e820.map[i].size);
+			res->name = "reserved RAM";
+			break;
+#endif
 		default:	res->name = "reserved";
 		}
 		res->start = e820.map[i].addr;
@@ -377,6 +411,16 @@ void __init e820_mark_nosave_regions(voi
 	}
 }
 
+static int __init e820_is_not_ram(int type)
+{
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+	return type != E820_RAM && type != E820_RESERVED_RAM &&
+		type != E820_RESERVED_RAM_FAILED;
+#else
+	return type != E820_RAM;
+#endif	
+}
+
 /*
  * Finds an active region in the address range from start_pfn to end_pfn and
  * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
@@ -395,11 +439,11 @@ static int __init e820_find_active_regio
 		return 0;
 
 	/* Check if max_pfn_mapped should be updated */
-	if (ei->type != E820_RAM && *ei_endpfn > max_pfn_mapped)
+	if (e820_is_not_ram(ei->type) && *ei_endpfn > max_pfn_mapped)
 		max_pfn_mapped = *ei_endpfn;
 
 	/* Skip if map is outside the node */
-	if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
+	if (e820_is_not_ram(ei->type) || *ei_endpfn <= start_pfn ||
 				    *ei_startpfn >= end_pfn)
 		return 0;
 
@@ -495,6 +539,14 @@ static void __init e820_print_map(char *
 		case E820_NVS:
 			printk(KERN_CONT "(ACPI NVS)\n");
 			break;
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		case E820_RESERVED_RAM:
+			printk(KERN_CONT "(reserved RAM)\n");
+			break;
+		case E820_RESERVED_RAM_FAILED:
+			printk(KERN_CONT "(reserved RAM failed)\n");
+			break;
+#endif
 		default:
 			printk(KERN_CONT "type %u\n", e820.map[i].type);
 			break;
@@ -724,9 +776,31 @@ static int __init copy_e820_map(struct e
 		u64 end = start + size;
 		u32 type = biosmap->type;
 
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		/* make space for two more low-prio types */
+		type += 2;
+#endif
+
 		/* Overflow in 64 bits? Ignore the memory map. */
 		if (start > end)
 			return -1;
+
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		if (type == E820_RAM) {
+			if (end <= __PHYSICAL_START) {
+				add_memory_region(start, size,
+						  E820_RESERVED_RAM);
+				continue;
+			}
+			if (start < __PHYSICAL_START) {
+				add_memory_region(start,
+						  __PHYSICAL_START-start,
+						  E820_RESERVED_RAM);
+				size -= __PHYSICAL_START-start;
+				start = __PHYSICAL_START;
+			}
+		}
+#endif
 
 		add_memory_region(start, size, type);
 	} while (biosmap++, --nr_map);
diff --git a/include/asm-x86/e820.h b/include/asm-x86/e820.h
--- a/include/asm-x86/e820.h
+++ b/include/asm-x86/e820.h
@@ -4,10 +4,19 @@
 #define E820MAX	128		/* number of entries in E820MAP */
 #define E820NR	0x1e8		/* # entries in E820MAP */
 
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+#define E820_RESERVED_RAM 1
+#define E820_RESERVED_RAM_FAILED 2
+#define E820_RAM	3
+#define E820_RESERVED	4
+#define E820_ACPI	5
+#define E820_NVS	6
+#else
 #define E820_RAM	1
 #define E820_RESERVED	2
 #define E820_ACPI	3
 #define E820_NVS	4
+#endif
 
 #ifndef __ASSEMBLY__
 struct e820entry {
diff --git a/include/asm-x86/page_64.h b/include/asm-x86/page_64.h
--- a/include/asm-x86/page_64.h
+++ b/include/asm-x86/page_64.h
@@ -29,6 +29,7 @@
 #define __PAGE_OFFSET           _AC(0xffff810000000000, UL)
 
 #define __PHYSICAL_START	CONFIG_PHYSICAL_START
+#define __PHYSICAL_OFFSET	(__PHYSICAL_START-0x200000)
 #define __KERNEL_ALIGN		0x200000
 
 /*
@@ -51,7 +52,7 @@
  * Kernel image size is limited to 512 MB (see level2_kernel_pgt in
  * arch/x86/kernel/head_64.S), and it is mapped here:
  */
-#define KERNEL_IMAGE_SIZE	(512 * 1024 * 1024)
+#define KERNEL_IMAGE_SIZE	(512 * 1024 * 1024 + __PHYSICAL_OFFSET)
 #define KERNEL_IMAGE_START	_AC(0xffffffff80000000, UL)
 
 #ifndef __ASSEMBLY__
diff --git a/include/asm-x86/pgtable_64.h b/include/asm-x86/pgtable_64.h
--- a/include/asm-x86/pgtable_64.h
+++ b/include/asm-x86/pgtable_64.h
@@ -145,7 +145,7 @@ static inline void native_pgd_clear(pgd_
 #define VMALLOC_START    _AC(0xffffc20000000000, UL)
 #define VMALLOC_END      _AC(0xffffe1ffffffffff, UL)
 #define VMEMMAP_START	 _AC(0xffffe20000000000, UL)
-#define MODULES_VADDR    _AC(0xffffffffa0000000, UL)
+#define MODULES_VADDR    (0xffffffffa0000000UL+__PHYSICAL_OFFSET)
 #define MODULES_END      _AC(0xfffffffffff00000, UL)
 #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
 
diff --git a/include/asm-x86/trampoline.h b/include/asm-x86/trampoline.h
--- a/include/asm-x86/trampoline.h
+++ b/include/asm-x86/trampoline.h
@@ -13,7 +13,11 @@ extern unsigned long init_rsp;
 extern unsigned long init_rsp;
 extern unsigned long initial_code;
 
+#ifndef CONFIG_RESERVE_PHYSICAL_START
 #define TRAMPOLINE_BASE 0x6000
+#else
+#define TRAMPOLINE_BASE 0x90000 /* move it next to 640k */
+#endif
 extern unsigned long setup_trampoline(void);
 
 #endif /* __ASSEMBLY__ */

[PATCH] reserved-ram kvm-userland patch

[Andrea Arcangeli]

This is the kvm-userland patch to use after applying the reserved-ram
patch to the host kernel. Bios must be rebuilt after applying the
patch, to do that just 'make bios'.

Then it's enough to pass '-reserved-ram' on the command line.

 4997 ?        Sl     2:56   3515  1544 4677235 1697028 47.9 /home/andrea/bin/x86_64/kvm/bin/qemu-system-x86_64 -hda tmp/virt
 5002 ?        Sl     3:23   4728  1544 4677235 1600980 45.2 /home/andrea/bin/x86_64/kvm/bin/qemu-system-x86_64 -hda tmp/virt
 5008 ?        Sl     2:39    239  1544 892127 15496  0.4 /home/andrea/bin/x86_64/kvm/bin/qemu-system-x86_64 -hda tmp/virtual

             total       used       free     shared    buffers     cached
Mem:       3540492    3525108      15384          0       1892      51896
-/+ buffers/cache:    3471320      69172
Swap:      5863684    3014072    2849612


eth0: no IPv6 routers present
loaded kvm module (kvm-70-399-g275f337)
apic write: bad size=1 fee00030
Ignoring de-assert INIT to vcpu 0
apic write: bad size=1 fee00030
Ignoring de-assert INIT to vcpu 0
Ignoring de-assert INIT to vcpu 0
Ignoring de-assert INIT to vcpu 0
kvm: emulating exchange as write
apic write: bad size=1 fee00030
Ignoring de-assert INIT to vcpu 0
Ignoring de-assert INIT to vcpu 0

You can see above 3 KVM guests, last one with -reserved-ram -m 512,
the first two with -m 3000. Host kernel has both mmu-notifier v18 and
-reserved-ram patch applied. KVM kernel has the pfn-mmio patch applied
plus my fix to export the reserved RAM through vma->fault, and the kvm
mmu notifier support for reliable and efficient swapping. All 3 guests
seems to work great together while system is 3G into swap. The
reserved-ram guest is almost responsive as if there would be no swap
of course (only the userland bits need to be paged in but all the
virtual ram remains in ram).

You can also see the RSS of the -reserved-ram task is only 15M which
is about the footprint of kvm userland (part of which are shared libs,
so it's actually much less).

Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>

diff --git a/bios/rombios.c b/bios/rombios.c
index 318de57..f93a6c6 100644
--- a/bios/rombios.c
+++ b/bios/rombios.c
@@ -4251,6 +4251,7 @@ int15_function32(regs, ES, DS, FLAGS)
   Bit32u  extra_lowbits_memory_size=0;
   Bit16u  CX,DX;
   Bit8u   extra_highbits_memory_size=0;
+  Bit32u  below_640_end;
 
 BX_DEBUG_INT15("int15 AX=%04x\n",regs.u.r16.ax);
 
@@ -4305,6 +4306,11 @@ ASM_END
          case 0x20: // coded by osmaker aka K.J.
             if(regs.u.r32.edx == 0x534D4150)
             {
+                below_640_end = inb_cmos(0x16);
+                below_640_end <<= 8;
+                below_640_end |= inb_cmos(0x15);
+                below_640_end *= 1024;
+
                 extended_memory_size = inb_cmos(0x35);
                 extended_memory_size <<= 8;
                 extended_memory_size |= inb_cmos(0x34);
@@ -4334,7 +4340,7 @@ ASM_END
                 {
                     case 0:
                         set_e820_range(ES, regs.u.r16.di,
-                                       0x0000000L, 0x0009fc00L, 0, 0, 1);
+                                       0x0000000L, below_640_end, 0, 0, 1);
                         regs.u.r32.ebx = 1;
                         regs.u.r32.eax = 0x534D4150;
                         regs.u.r32.ecx = 0x14;
@@ -4343,7 +4349,7 @@ ASM_END
                         break;
                     case 1:
                         set_e820_range(ES, regs.u.r16.di,
-                                       0x0009fc00L, 0x000a0000L, 0, 0, 2);
+                                       below_640_end, 0x000a0000L, 0, 0, 2);
                         regs.u.r32.ebx = 2;
                         regs.u.r32.eax = 0x534D4150;
                         regs.u.r32.ecx = 0x14;
diff --git a/qemu/hw/pc.c b/qemu/hw/pc.c
index 42c2687..c6a21d5 100644
--- a/qemu/hw/pc.c
+++ b/qemu/hw/pc.c
@@ -235,6 +235,8 @@ static void cmos_init(ram_addr_t ram_size, ram_addr_t above_4g_mem_size,
 
     /* memory size */
     val = 640; /* base memory in K */
+    if (reserved_ram)
+	    val = reserved[1] / 1024;
     rtc_set_memory(s, 0x15, val);
     rtc_set_memory(s, 0x16, val >> 8);
 
diff --git a/qemu/pc-bios/bios.bin b/qemu/pc-bios/bios.bin
index 3e5d96a..c9c94e6 100644
Binary files a/qemu/pc-bios/bios.bin and b/qemu/pc-bios/bios.bin differ
diff --git a/qemu/sysemu.h b/qemu/sysemu.h
index 97d73e9..964fee4 100644
--- a/qemu/sysemu.h
+++ b/qemu/sysemu.h
@@ -102,6 +102,8 @@ extern int autostart;
 extern int old_param;
 extern int hpagesize;
 extern const char *bootp_filename;
+extern int reserved_ram;
+extern int64_t reserved[4];
 
 
 #ifdef USE_KQEMU
diff --git a/qemu/vl.c b/qemu/vl.c
index f573dce..3ce2f2a 100644
--- a/qemu/vl.c
+++ b/qemu/vl.c
@@ -235,6 +235,8 @@ int time_drift_fix = 0;
 unsigned int kvm_shadow_memory = 0;
 const char *mem_path = NULL;
 int hpagesize = 0;
+int reserved_ram = 0;
+int64_t reserved[4];
 const char *cpu_vendor_string;
 #ifdef TARGET_ARM
 int old_param = 0;
@@ -7817,6 +7819,7 @@ static void help(int exitcode)
            "-clock          force the use of the given methods for timer alarm.\n"
            "                To see what timers are available use -clock ?\n"
            "-startdate      select initial date of the clock\n"
+           "-reserved-ram   use reserved RAM in /proc/iomem with spte identity mapping\n"
            "\n"
            "During emulation, the following keys are useful:\n"
            "ctrl-alt-f      toggle full screen\n"
@@ -7932,6 +7935,7 @@ enum {
     QEMU_OPTION_tdf,
     QEMU_OPTION_kvm_shadow_memory,
     QEMU_OPTION_mempath,
+    QEMU_OPTION_reserved_ram,
 };
 
 typedef struct QEMUOption {
@@ -8059,6 +8063,7 @@ const QEMUOption qemu_options[] = {
     { "startdate", HAS_ARG, QEMU_OPTION_startdate },
     { "tb-size", HAS_ARG, QEMU_OPTION_tb_size },
     { "mem-path", HAS_ARG, QEMU_OPTION_mempath },
+    { "reserved-ram", 0, QEMU_OPTION_reserved_ram },
     { NULL },
 };
 
@@ -8276,6 +8281,77 @@ static int gethugepagesize(void)
     return hugepagesize;
 }
 
+static int find_reserved_ram(int64_t *_start, int64_t *_end,
+			     unsigned long below, unsigned long above,
+			     unsigned long min_size)
+{
+    int ret, fd;
+    char buf[4096];
+    char *needle = "reserved RAM\n";
+    char *size, *curr;
+    int64_t start, end;
+
+    fd = open("/proc/iomem", O_RDONLY);
+    if (fd < 0) {
+	perror("open");
+	exit(0);
+    }
+
+    ret = read(fd, buf, sizeof(buf)-1);
+    if (ret < 0) {
+	perror("read");
+	exit(0);
+    }
+    buf[ret] = 0;
+
+    size = buf;
+    while (1) {
+	    size = strstr(size, needle);
+	    if (!size)
+		    return 0;
+	    size += strlen(needle);
+	    curr = size - strlen(needle) - 20;
+	    start = strtoll(curr, &curr, 16);
+	    end = strtoll(curr+1, NULL, 16);
+	    if ((!above || start >= above) && (!below || end <= below) &&
+		(!min_size || end-start >= min_size)) {
+		    *_start = start;
+		    *_end = end+1;
+		    return 1;
+	    }
+    }
+}
+
+static void init_reserved_ram(void)
+{
+	if (find_reserved_ram(&reserved[0], &reserved[1],
+			      640*1024, 0, 500*1024) &&
+	    find_reserved_ram(&reserved[2], &reserved[3],
+			      0, 1024*1024, 1024*1024)) {
+		reserved_ram = 1;
+		if (reserved[0] != 4096) {
+			fprintf(stderr,
+				"strange host ram layout\n");
+			exit(1);
+		}
+		if (reserved[2] != 1024*1024) {
+			fprintf(stderr,
+				"strange host ram layout\n");
+			exit(1);
+		}
+		if (reserved[3] < ram_size) {
+			fprintf(stderr,
+				"not enough host reserved ram, decrease -m\n");
+			exit(1);
+		}
+		reserved[1] &= TARGET_PAGE_MASK;
+	} else {
+		fprintf(stderr,
+			"host reserved ram not found\n");
+		exit(1);
+	}
+}
+
 void *alloc_mem_area(unsigned long memory, const char *path)
 {
     char *filename;
@@ -8322,10 +8398,43 @@ void *qemu_alloc_physram(unsigned long memory)
 {
     void *area = NULL;
 
-    if (mem_path)
+    if (!area && mem_path)
 	area = alloc_mem_area(memory, mem_path);
-    if (!area)
+    if (!area) {
 	area = qemu_vmalloc(memory);
+	if (reserved_ram) {
+	    int fd;
+	    if (memory < reserved[2]) {
+		printf("memory < reserved[2]\n");
+		return NULL;
+	    }
+	    fd = open("/dev/mem", O_RDWR);
+	    if (fd < 0) {
+		perror("reserved_ram requires access to /dev/mem");
+		return NULL;
+	    }
+	    if (mmap((char *)area+reserved[0],
+		reserved[1]-reserved[0],
+		     PROT_READ|PROT_WRITE, MAP_SHARED|MAP_FIXED,
+		     fd, 0) == MAP_FAILED) {
+		    perror("reserved_ram mmap failed on /dev/mem");
+		    return NULL;
+	    }
+	    bzero((char *)area+reserved[0], reserved[1]-reserved[0]);
+	    if (mmap((char *)area+reserved[2],
+		     ram_size-reserved[2],
+		     PROT_READ|PROT_WRITE, MAP_SHARED|MAP_FIXED,
+		     fd, reserved[2]) == MAP_FAILED) {
+		    perror("reserved_ram mmap failed on /dev/mem");
+		    return NULL;
+	    }
+	    bzero((char *)area+reserved[2], ram_size-reserved[2]);
+	    if (close(fd) < 0) {
+		    perror("/dev/mem");
+		    return NULL;
+	    }
+	}
+    }
 
     return area;
 }
@@ -8962,6 +9071,9 @@ int main(int argc, char **argv)
             case QEMU_OPTION_mempath:
 		mem_path = optarg;
 		break;
+            case QEMU_OPTION_reserved_ram:
+		init_reserved_ram();
+		break;
             case QEMU_OPTION_name:
                 qemu_name = optarg;
                 break;

[PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andrea Arcangeli]

From: Andrea Arcangeli <andrea@qumranet.com>

The "reserved RAM" can be mapped by virtualization software with
/dev/mem to create a 1:1 mapping between guest physical (bus) address
and host physical (bus) address. This will allow pci passthrough with
DMA for the guest using the ram with the 1:1 mapping. The only detail
to take care of is the ram marked "reserved RAM failed". The
virtualization software must create for the guest an e820 map that
only includes the "reserved RAM" regions but if the guest touches
memory with guest physical address in the "reserved RAM failed" ranges
(linux guest will do that even if the ram isn't present in the e820
map), it should provide that as ram and map it with a non linear
mapping. This should allow any linux kernel to run fine and hopefully
any other OS too.

svm ~ # cat /proc/iomem |head -n 20
00000000-00000fff : reserved RAM failed
00001000-00005fff : reserved RAM
00006000-00007fff : reserved RAM failed
00008000-0009efff : reserved RAM
0009f000-0009ffff : reserved
000cd600-000cffff : pnp 00:0d
000f0000-000fffff : reserved
00100000-0fffffff : reserved RAM
10000000-3dedffff : System RAM
  10000000-10329ab2 : Kernel code
  10329ab3-104933e7 : Kernel data
  104f5000-10558e67 : Kernel bss
3dee0000-3dee2fff : ACPI Non-volatile Storage
3dee3000-3deeffff : ACPI Tables
3def0000-3defffff : reserved
3dff0000-3ffeffff : pnp 00:0d
e0000000-efffffff : reserved
fa000000-fbffffff : PCI Bus #01
  fa000000-fbffffff : 0000:01:05.0
fda00000-fdbfffff : PCI Bus #01
svm ~ # hexdump /dev/mem | grep -C2 'cccc cccc cccc cccc'
00007e0 0000 0000 0000 0000 0000 0000 0000 0000
*
0001000 cccc cccc cccc cccc cccc cccc cccc cccc
*
0006000 a5a5 a5a5 8ec8 8ed8 8ec0 66d0 06c7 0000
--
*
0007ff0 0000 0000 0000 0000 3063 1000 0000 0000
0008000 cccc cccc cccc cccc cccc cccc cccc cccc
*
009f000 0002 0000 0000 0000 0000 0000 0000 0000
--
00fffe0 6000 3c03 45e7 0184 0500 0082 01c0 0223
00ffff0 5bea 00e0 31f0 2f32 3931 302f 0037 12fc
0100000 cccc cccc cccc cccc cccc cccc cccc cccc
*
10000000 8d48 f92d ffff 48ff ed81 0000 1000 8948
^C
svm ~ #

Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
---

This is a port to current linux-2.6.git of the previous reserved-ram
patch. Let me know if there's a chance to get this acked and
included. Anything that isn't at compile time would require much
bigger changes just to parse the command line at 16bit realmode time
to know where to relocate the kernel dynamically. Because 1:1 is a
corner case feature required only by some users, this is the minimal
intrusive approach. This also has some limits as it can't reserve more
than 1g, and with a few more changes 2g but this is ok for a long time
as the virtualized 1:1 guest doesn't need to be huge, just a desktop.

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1276,8 +1276,36 @@ config CRASH_DUMP
 	  (CONFIG_RELOCATABLE=y).
 	  For more details see Documentation/kdump/kdump.txt
 
+config RESERVE_PHYSICAL_START
+	bool "Reserve all RAM below PHYSICAL_START (EXPERIMENTAL)"
+	depends on !RELOCATABLE && X86_64
+	help
+	  This makes the kernel use only RAM above __PHYSICAL_START.
+	  All memory below __PHYSICAL_START will be left unused and
+	  marked as "reserved RAM" in /proc/iomem. The few special
+	  pages that can't be relocated at addresses above
+	  __PHYSICAL_START and that can't be guaranteed to be unused
+	  by the running kernel will be marked "reserved RAM failed"
+	  in /proc/iomem. Those may or may be not used by the kernel
+	  (for example SMP trampoline pages would only be used if
+	  CPU hotplug is enabled).
+
+	  The "reserved RAM" can be mapped by virtualization software
+	  with /dev/mem to create a 1:1 mapping between guest physical
+	  (bus) address and host physical (bus) address. This will
+	  allow PCI passthrough with DMA for the guest using the RAM
+	  with the 1:1 mapping. The only detail to take care of is the
+	  RAM marked "reserved RAM failed". The virtualization
+	  software must create for the guest an e820 map that only
+	  includes the "reserved RAM" regions but if the guest touches
+	  memory with guest physical address in the "reserved RAM
+	  failed" ranges (Linux guest will do that even if the RAM
+	  isn't present in the e820 map), it should provide that as
+	  RAM and map it with a non-linear mapping. This should allow
+	  any Linux kernel to run fine and hopefully any other OS too.
+
 config PHYSICAL_START
-	hex "Physical address where the kernel is loaded" if (EMBEDDED || CRASH_DUMP)
+	hex "Physical address where the kernel is loaded" if (EMBEDDED || CRASH_DUMP || RESERVE_PHYSICAL_START)
 	default "0x1000000" if X86_NUMAQ
 	default "0x200000" if X86_64
 	default "0x100000"
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
--- a/arch/x86/kernel/e820.c
+++ b/arch/x86/kernel/e820.c
@@ -148,6 +148,14 @@ void __init e820_print_map(char *who)
 		case E820_NVS:
 			printk(KERN_CONT "(ACPI NVS)\n");
 			break;
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		case E820_RESERVED_RAM:
+			printk(KERN_CONT "(reserved RAM)\n");
+			break;
+		case E820_RESERVED_RAM_FAILED:
+			printk(KERN_CONT "(reserved RAM failed)\n");
+			break;
+#endif
 		default:
 			printk(KERN_CONT "type %u\n", e820.map[i].type);
 			break;
@@ -384,10 +392,28 @@ static int __init __append_e820_map(stru
 		u64 end = start + size;
 		u32 type = biosmap->type;
 
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		/* make space for two more low-prio types */
+		type += 2;
+#endif
+
 		/* Overflow in 64 bits? Ignore the memory map. */
 		if (start > end)
 			return -1;
 
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		if (type == E820_RAM) {
+			if (end <= __PHYSICAL_START)
+				type = E820_RESERVED_RAM;
+			else if (start < __PHYSICAL_START) {
+				e820_add_region(start,
+						__PHYSICAL_START-start,
+						E820_RESERVED_RAM);
+				size -= __PHYSICAL_START-start;
+				start = __PHYSICAL_START;
+			}
+		}
+#endif
 		e820_add_region(start, size, type);
 
 		biosmap++;
@@ -893,7 +919,35 @@ void __init early_res_to_bootmem(u64 sta
 			final_start, final_end);
 		reserve_bootmem_generic(final_start, final_end - final_start,
 				BOOTMEM_DEFAULT);
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		if (r->start < __PHYSICAL_START)
+			e820_add_region(r->start, r->end - r->start,
+					E820_RESERVED_RAM_FAILED);
+#endif			
 	}
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+	/* solve E820_RESERVED_RAM vs E820_RESERVED_RAM_FAILED conflicts */
+	update_e820();
+
+	/* now reserve E820_RESERVED_RAM */
+	for (i = 0; i < e820.nr_map; i++) {
+		struct e820entry *ei = &e820.map[i];
+
+		if (ei->type != E820_RESERVED_RAM)
+			continue;
+		final_start = max(start, (u64) ei->addr);
+		final_end = min(end, (u64) (ei->addr + ei->size));
+		if (final_start >= final_end)
+			continue;
+		if (reserve_bootmem_generic(final_start,
+					    final_end - final_start,
+					    BOOTMEM_DEFAULT))
+			printk(KERN_ERR "reserved physical start failure");
+		else
+			printk(KERN_INFO " bootmem reserved RAM: [%Lx-%Lx]\n",
+			       final_start, final_end - 1);
+	}
+#endif
 }
 
 /* Check for already reserved areas */
@@ -1095,6 +1149,17 @@ unsigned long __init e820_end_of_low_ram
 {
 	return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
 }
+
+static int __init e820_is_not_ram(int type)
+{
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+	return type != E820_RAM && type != E820_RESERVED_RAM &&
+		type != E820_RESERVED_RAM_FAILED;
+#else
+	return type != E820_RAM;
+#endif	
+}
+
 /*
  * Finds an active region in the address range from start_pfn to last_pfn and
  * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
@@ -1115,8 +1180,8 @@ int __init e820_find_active_region(const
 		return 0;
 
 	/* Skip if map is outside the node */
-	if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
-				    *ei_startpfn >= last_pfn)
+	if (e820_is_not_ram(ei->type) || *ei_endpfn <= start_pfn ||
+	    *ei_startpfn >= last_pfn)
 		return 0;
 
 	/* Check for overlaps */
@@ -1260,6 +1325,10 @@ static inline const char *e820_type_to_s
 	case E820_RAM:	return "System RAM";
 	case E820_ACPI:	return "ACPI Tables";
 	case E820_NVS:	return "ACPI Non-volatile Storage";
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+	case E820_RESERVED_RAM_FAILED: return "reserved RAM failed";
+	case E820_RESERVED_RAM: return "reserved RAM";
+#endif
 	default:	return "reserved";
 	}
 }
@@ -1289,6 +1358,12 @@ void __init e820_reserve_resources(void)
 		res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 		insert_resource(&iomem_resource, res);
 		res++;
+
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+		if (i == E820_RESERVED_RAM)
+			memset(__va(e820.map[i].addr),
+			       POISON_FREE_INITMEM, e820.map[i].size);
+#endif
 	}
 
 	for (i = 0; i < e820_saved.nr_map; i++) {
diff --git a/include/asm-x86/e820.h b/include/asm-x86/e820.h
--- a/include/asm-x86/e820.h
+++ b/include/asm-x86/e820.h
@@ -39,10 +39,19 @@
 
 #define E820NR	0x1e8		/* # entries in E820MAP */
 
+#ifdef CONFIG_RESERVE_PHYSICAL_START
+#define E820_RESERVED_RAM 1
+#define E820_RESERVED_RAM_FAILED 2
+#define E820_RAM	3
+#define E820_RESERVED	4
+#define E820_ACPI	5
+#define E820_NVS	6
+#else
 #define E820_RAM	1
 #define E820_RESERVED	2
 #define E820_ACPI	3
 #define E820_NVS	4
+#endif
 
 /* reserved RAM used by kernel itself */
 #define E820_RESERVED_KERN        128
diff --git a/include/asm-x86/page_64.h b/include/asm-x86/page_64.h
--- a/include/asm-x86/page_64.h
+++ b/include/asm-x86/page_64.h
@@ -35,6 +35,7 @@
 #define __PAGE_OFFSET           _AC(0xffff880000000000, UL)
 
 #define __PHYSICAL_START	CONFIG_PHYSICAL_START
+#define __PHYSICAL_OFFSET	(__PHYSICAL_START-0x200000)
 #define __KERNEL_ALIGN		0x200000
 
 /*
@@ -57,7 +58,7 @@
  * Kernel image size is limited to 512 MB (see level2_kernel_pgt in
  * arch/x86/kernel/head_64.S), and it is mapped here:
  */
-#define KERNEL_IMAGE_SIZE	(512 * 1024 * 1024)
+#define KERNEL_IMAGE_SIZE	(512 * 1024 * 1024 + __PHYSICAL_OFFSET)
 #define KERNEL_IMAGE_START	_AC(0xffffffff80000000, UL)
 
 #ifndef __ASSEMBLY__
diff --git a/include/asm-x86/pgtable_64.h b/include/asm-x86/pgtable_64.h
--- a/include/asm-x86/pgtable_64.h
+++ b/include/asm-x86/pgtable_64.h
@@ -150,7 +150,7 @@ static inline void native_pgd_clear(pgd_
 #define VMALLOC_START    _AC(0xffffc20000000000, UL)
 #define VMALLOC_END      _AC(0xffffe1ffffffffff, UL)
 #define VMEMMAP_START	 _AC(0xffffe20000000000, UL)
-#define MODULES_VADDR    _AC(0xffffffffa0000000, UL)
+#define MODULES_VADDR    (0xffffffffa0000000UL+__PHYSICAL_OFFSET)
 #define MODULES_END      _AC(0xfffffffffff00000, UL)
 #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
 
diff --git a/include/asm-x86/trampoline.h b/include/asm-x86/trampoline.h
--- a/include/asm-x86/trampoline.h
+++ b/include/asm-x86/trampoline.h
@@ -13,7 +13,11 @@ extern unsigned long init_rsp;
 extern unsigned long init_rsp;
 extern unsigned long initial_code;
 
+#ifndef CONFIG_RESERVE_PHYSICAL_START
 #define TRAMPOLINE_BASE 0x6000
+#else
+#define TRAMPOLINE_BASE 0x90000 /* move it next to 640k */
+#endif
 extern unsigned long setup_trampoline(void);
 
 #endif /* __ASSEMBLY__ */

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andi Kleen]

> This is a port to current linux-2.6.git of the previous reserved-ram
> patch. Let me know if there's a chance to get this acked and
> included. Anything that isn't at compile time would require much

I still think runtime would be far better. Nobody really wants
a proliferation of more weird special kernel images.

> bigger changes just to parse the command line at 16bit realmode time

You could always do it with kexec if you think 16bit real mode is
too hard.

-Andi

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andrea Arcangeli]

On Tue, Jul 29, 2008 at 02:43:17PM +0200, Andi Kleen wrote:
> > This is a port to current linux-2.6.git of the previous reserved-ram
> > patch. Let me know if there's a chance to get this acked and
> > included. Anything that isn't at compile time would require much
> 
> I still think runtime would be far better. Nobody really wants
> a proliferation of more weird special kernel images.

Not for the usage we're interested about but surely this would prevent
distro to take advantage of the feature. The question is if distro
need to take advantage of the feature in the first place instead of
sticking with VT-d. 1:1 isn't secure virtualization as the guest must
be trusted so it's not necessarily a good model to deploy to users
that don't know exactly what they're doing.

> > bigger changes just to parse the command line at 16bit realmode time
> 
> You could always do it with kexec if you think 16bit real mode is
> too hard.

It's not too hard, but it'll add bloat to the 16 bit part of the boot
in the bzImage. It's likely simpler than kexec and surely more
user-friendly to setup for the end user.

In any case, my patch does the needed bits with regard to the e820
map. An incremental patch can add the parsing of the booatloader and
switch the Kconfig dependency from PHYSICAL_START to RELOCATABLE. The
e820 file will then have to replace the __PHYSICAL_START define with
something else and that's all.

I mean it's not entirely backwards to provide a compile time smaller
and simpler approach initially, and then to go where you want to go
incrementally later if we're sure there's enough userbase needing 1:1.

I'm not so interested to go there right now, because while this code
is useful right now because the majority of systems out there lacks
VT-d/iommu, I suspect this code could be nuked in the long
run when all systems will ship with that, which is why I kept it all
under #ifdef, and the changes to the other files outside ifdef are
bugfixes needed if you want to kexec-relocate above 40m or so that
should be kept.

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andi Kleen]

> I'm not so interested to go there right now, because while this code
> is useful right now because the majority of systems out there lacks
> VT-d/iommu, I suspect this code could be nuked in the long
> run when all systems will ship with that, which is why I kept it all

Actually at least on Intel platforms and if you exclude the lowest end
VT-d is shipping universally for quite some time now. If you
buy a Intel box today or bought it in the last year the chances are pretty 
high that it has VT-d support.

> under #ifdef, and the changes to the other files outside ifdef are
> bugfixes needed if you want to kexec-relocate above 40m or so that
> should be kept.

You should split that out then into a separate patch.

-Andi

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Amit Shah]

* On Tuesday 29 July 2008 18:47:35 Andi Kleen wrote:
> > I'm not so interested to go there right now, because while this code
> > is useful right now because the majority of systems out there lacks
> > VT-d/iommu, I suspect this code could be nuked in the long
> > run when all systems will ship with that, which is why I kept it all
>
> Actually at least on Intel platforms and if you exclude the lowest end
> VT-d is shipping universally for quite some time now. If you
> buy a Intel box today or bought it in the last year the chances are pretty
> high that it has VT-d support.

I think you mean VT-x, which is virtualization extensions for the x86 
architecture. VT-d is virtualization extensions for devices (IOMMU).

Amit

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andi Kleen]

On Wed, Jul 30, 2008 at 11:50:43AM +0530, Amit Shah wrote:
> * On Tuesday 29 July 2008 18:47:35 Andi Kleen wrote:
> > > I'm not so interested to go there right now, because while this code
> > > is useful right now because the majority of systems out there lacks
> > > VT-d/iommu, I suspect this code could be nuked in the long
> > > run when all systems will ship with that, which is why I kept it all
> >
> > Actually at least on Intel platforms and if you exclude the lowest end
> > VT-d is shipping universally for quite some time now. If you
> > buy a Intel box today or bought it in the last year the chances are pretty
> > high that it has VT-d support.
> 
> I think you mean VT-x, which is virtualization extensions for the x86 
> architecture. VT-d is virtualization extensions for devices (IOMMU).

No I really mean VT-d. The modern not very lowend Intel IOHubs all have it.

-Andi

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andrea Arcangeli]

On Wed, Jul 30, 2008 at 11:50:43AM +0530, Amit Shah wrote:
> * On Tuesday 29 July 2008 18:47:35 Andi Kleen wrote:
> > > I'm not so interested to go there right now, because while this code
> > > is useful right now because the majority of systems out there lacks
> > > VT-d/iommu, I suspect this code could be nuked in the long
> > > run when all systems will ship with that, which is why I kept it all
> >
> > Actually at least on Intel platforms and if you exclude the lowest end
> > VT-d is shipping universally for quite some time now. If you
> > buy a Intel box today or bought it in the last year the chances are pretty
> > high that it has VT-d support.
> 
> I think you mean VT-x, which is virtualization extensions for the x86 
> architecture. VT-d is virtualization extensions for devices (IOMMU).

I think Andi understood VT-d right but even if he was right that every
reader of this email that is buying a new VT-x system today is also
almost guaranteed to get a VT-d motherboard (which I disagree unless
you buy some really expensive toy), there are current large
installations of VT-x systems that lacks VT-d and that with recent
current dual/quadcore cpus are very fast and will be used for the next
couple of years and they will not upgrade just the motherboard to use
pci-passthrough.

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Dor Laor]

Andrea Arcangeli wrote:
> On Wed, Jul 30, 2008 at 11:50:43AM +0530, Amit Shah wrote:
>   
>> * On Tuesday 29 July 2008 18:47:35 Andi Kleen wrote:
>>     
>>>> I'm not so interested to go there right now, because while this code
>>>> is useful right now because the majority of systems out there lacks
>>>> VT-d/iommu, I suspect this code could be nuked in the long
>>>> run when all systems will ship with that, which is why I kept it all
>>>>         
>>> Actually at least on Intel platforms and if you exclude the lowest end
>>> VT-d is shipping universally for quite some time now. If you
>>> buy a Intel box today or bought it in the last year the chances are pretty
>>> high that it has VT-d support.
>>>       
>> I think you mean VT-x, which is virtualization extensions for the x86 
>> architecture. VT-d is virtualization extensions for devices (IOMMU).
>>     
>
> I think Andi understood VT-d right but even if he was right that every
> reader of this email that is buying a new VT-x system today is also
> almost guaranteed to get a VT-d motherboard (which I disagree unless
> you buy some really expensive toy), there are current large
> installations of VT-x systems that lacks VT-d and that with recent
> current dual/quadcore cpus are very fast and will be used for the next
> couple of years and they will not upgrade just the motherboard to use
> pci-passthrough.
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   

In addition KVM is used in embedded too and things are slower there, we 
know of a specific use case (production) that demands
1:1 mapping and can't use VT-d

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[Andrea Arcangeli]

On Wed, Jul 30, 2008 at 05:16:06PM +0300, Dor Laor wrote:
> In addition KVM is used in embedded too and things are slower there, we 
> know of a specific use case (production) that demands
> 1:1 mapping and can't use VT-d

Since you mentioned this ;), I take opportunity to add that those
embedded usages are the ones that are totally fine with the compile
time passthrough-guest-ram decision, instead of a boot time
decision. Those host kernels will likely have RT patches (KVM works
great with preempt-RT indeed) and in turn the compile time ram
selection is the least of their problems as you can imagine ;). So you
can see my patch as an embedded-build option, similar to "Configure
standard kernel features (for small systems)" and no distro is
shipping new kernels with that feature on either.

Than if we decide 1:1 should have larger userbase instead of only the
people that knows what they're doing (i.e. 1:1 guest can destroy
linux-hypervisor) we can always add a bit of strtol parsing to 16bit
kernelloader.

Re: [PATCH] reserved-ram for pci-passthrough without VT-d capable hardware

[FUJITA Tomonori]

On Wed, 30 Jul 2008 15:58:46 +0200
Andrea Arcangeli <andrea@qumranet.com> wrote:

> On Wed, Jul 30, 2008 at 11:50:43AM +0530, Amit Shah wrote:
> > * On Tuesday 29 July 2008 18:47:35 Andi Kleen wrote:
> > > > I'm not so interested to go there right now, because while this code
> > > > is useful right now because the majority of systems out there lacks
> > > > VT-d/iommu, I suspect this code could be nuked in the long
> > > > run when all systems will ship with that, which is why I kept it all
> > >
> > > Actually at least on Intel platforms and if you exclude the lowest end
> > > VT-d is shipping universally for quite some time now. If you
> > > buy a Intel box today or bought it in the last year the chances are pretty
> > > high that it has VT-d support.
> > 
> > I think you mean VT-x, which is virtualization extensions for the x86 
> > architecture. VT-d is virtualization extensions for devices (IOMMU).
> 
> I think Andi understood VT-d right but even if he was right that every
> reader of this email that is buying a new VT-x system today is also
> almost guaranteed to get a VT-d motherboard (which I disagree unless
> you buy some really expensive toy), there are current large
> installations of VT-x systems that lacks VT-d and that with recent
> current dual/quadcore cpus are very fast and will be used for the next
> couple of years and they will not upgrade just the motherboard to use
> pci-passthrough.

Today, very inexpensive desktops (for example, Dell OptiPlex 755) have
VT-d support.