[patch 0/3] Protect crashkernel against BSS overlap

[patch 0/3] Protect crashkernel against BSS overlap

[Bernhard Walle]

I observed the problem that even when you choose the default 16M as
crashkernel base address and the kernel is very big, the reserved area may
overlap with the kernel BSS. Currently, this is not checked at runtime, so the
kernel just crashes when you load the panic kernel in the sys_kexec call.

This three patches check this at runtime. The patches are against current git,
but with the patches

    extended-crashkernel-command-line.patch
    extended-crashkernel-command-line-update.patch
    extended-crashkernel-command-line-comment-fix.patch
    extended-crashkernel-command-line-improve-error-handling-in-parse_crashkernel_mem.patch
    use-extended-crashkernel-command-line-on-i386.patch
    use-extended-crashkernel-command-line-on-i386-update.patch
    use-extended-crashkernel-command-line-on-x86_64.patch
    use-extended-crashkernel-command-line-on-x86_64-update.patch
    use-extended-crashkernel-command-line-on-ia64.patch
    use-extended-crashkernel-command-line-on-ia64-fix.patch
    use-extended-crashkernel-command-line-on-ia64-update.patch
    use-extended-crashkernel-command-line-on-ppc64.patch
    use-extended-crashkernel-command-line-on-ppc64-update.patch
    use-extended-crashkernel-command-line-on-sh.patch
    use-extended-crashkernel-command-line-on-sh-update.patch

from -mm tree applied since they are marked to be merged in 2.6.24.

I know that the implementation of both patches is only x86 (i386 and x86-64),
but if you agree that it's the way to go, I'll modify the patch for all
architectures.


Signed-off-by: Bernhard Walle <bwalle@suse.de>

-- 

[patch 1/3] Add BSS to resource tree

[Bernhard Walle]

[-- Attachment #1: add-bss-resource --]
[-- Type: text/plain, Size: 8782 bytes --]

This patch adds the BSS to the resource tree just as kernel text and kernel
data are in the resource tree. The main reason behind this is to avoid
crashkernel reservation in that area.

While it's not strictly necessary to have the BSS in the resource tree
(the actual collision detection is done in the reserve_bootmem() function
before), the usage of the BSS resource should be presented to the user
in /proc/iomem just as Kernel data and Kernel code.

Note: The patch currently is only implemented for x86 and ia64 (because
efi_initialize_iomem_resources() has the same signature on i386 and
ia64).


Signed-off-by: Bernhard Walle <bwalle@suse.de>

---
 arch/ia64/kernel/efi.c     |   10 ++++++----
 arch/ia64/kernel/setup.c   |    9 ++++++++-
 arch/x86/kernel/e820_32.c  |   18 +++++++++++++-----
 arch/x86/kernel/e820_64.c  |    3 ++-
 arch/x86/kernel/efi_32.c   |   11 +++++++----
 arch/x86/kernel/setup_32.c |    4 ++++
 arch/x86/kernel/setup_64.c |    9 +++++++++
 include/linux/efi.h        |    3 +--
 8 files changed, 50 insertions(+), 17 deletions(-)

--- a/arch/ia64/kernel/efi.c
+++ b/arch/ia64/kernel/efi.c
@@ -41,6 +41,8 @@
 
 extern efi_status_t efi_call_phys (void *, ...);
 
+extern struct resource code_resource, data_resource, bss_resource;
+
 struct efi efi;
 EXPORT_SYMBOL(efi);
 static efi_runtime_services_t *runtime;
@@ -1089,8 +1091,7 @@ efi_memmap_init(unsigned long *s, unsign
 }
 
 void
-efi_initialize_iomem_resources(struct resource *code_resource,
-			       struct resource *data_resource)
+efi_initialize_iomem_resources(void)
 {
 	struct resource *res;
 	void *efi_map_start, *efi_map_end, *p;
@@ -1169,8 +1170,9 @@ efi_initialize_iomem_resources(struct re
 			 * kernel data so we try it repeatedly and
 			 * let the resource manager test it.
 			 */
-			insert_resource(res, code_resource);
-			insert_resource(res, data_resource);
+			insert_resource(res, &code_resource);
+			insert_resource(res, &data_resource);
+			insert_resource(res, &bss_resource);
 #ifdef CONFIG_KEXEC
                         insert_resource(res, &efi_memmap_res);
                         insert_resource(res, &boot_param_res);
--- a/arch/ia64/kernel/setup.c
+++ b/arch/ia64/kernel/setup.c
@@ -90,6 +90,11 @@ static struct resource code_resource = {
 	.name	= "Kernel code",
 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
 };
+
+static struct resource bss_resource = {
+	.name	= "Kernel bss",
+	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
+};
 extern char _text[], _end[], _etext[];
 
 unsigned long ia64_max_cacheline_size;
@@ -201,7 +206,9 @@ static int __init register_memory(void)
 	code_resource.end   = ia64_tpa(_etext) - 1;
 	data_resource.start = ia64_tpa(_etext);
 	data_resource.end   = ia64_tpa(_end) - 1;
-	efi_initialize_iomem_resources(&code_resource, &data_resource);
+	bss_resource.start  = ia64_tpa(__bss_start);
+	bss_resource.end    = ia64_tpa(__bss_stop) - 1;
+	efi_initialize_iomem_resources();
 
 	return 0;
 }
--- a/arch/x86/kernel/e820_32.c
+++ b/arch/x86/kernel/e820_32.c
@@ -51,6 +51,13 @@ struct resource code_resource = {
 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
 };
 
+struct resource bss_resource = {
+	.name	= "Kernel bss",
+	.start	= 0,
+	.end	= 0,
+	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
+};
+
 static struct resource system_rom_resource = {
 	.name	= "System ROM",
 	.start	= 0xf0000,
@@ -254,7 +261,7 @@ static void __init probe_roms(void)
  * and also for regions reported as reserved by the e820.
  */
 static void __init
-legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
+legacy_init_iomem_resources(void)
 {
 	int i;
 
@@ -285,8 +292,9 @@ legacy_init_iomem_resources(struct resou
 			 *  so we try it repeatedly and let the resource manager
 			 *  test it.
 			 */
-			request_resource(res, code_resource);
-			request_resource(res, data_resource);
+			request_resource(res, &code_resource);
+			request_resource(res, &data_resource);
+			request_resource(res, &bss_resource);
 #ifdef CONFIG_KEXEC
 			if (crashk_res.start != crashk_res.end)
 				request_resource(res, &crashk_res);
@@ -307,9 +315,9 @@ static int __init request_standard_resou
 
 	printk("Setting up standard PCI resources\n");
 	if (efi_enabled)
-		efi_initialize_iomem_resources(&code_resource, &data_resource);
+		efi_initialize_iomem_resources();
 	else
-		legacy_init_iomem_resources(&code_resource, &data_resource);
+		legacy_init_iomem_resources();
 
 	/* EFI systems may still have VGA */
 	request_resource(&iomem_resource, &video_ram_resource);
--- a/arch/x86/kernel/e820_64.c
+++ b/arch/x86/kernel/e820_64.c
@@ -47,7 +47,7 @@ unsigned long end_pfn_map; 
  */
 static unsigned long __initdata end_user_pfn = MAXMEM>>PAGE_SHIFT;
 
-extern struct resource code_resource, data_resource;
+extern struct resource code_resource, data_resource, bss_resource;
 
 /* Check for some hardcoded bad areas that early boot is not allowed to touch */ 
 static inline int bad_addr(unsigned long *addrp, unsigned long size)
@@ -220,6 +220,7 @@ void __init e820_reserve_resources(void)
 			 */
 			request_resource(res, &code_resource);
 			request_resource(res, &data_resource);
+			request_resource(res, &bss_resource);
 #ifdef CONFIG_KEXEC
 			if (crashk_res.start != crashk_res.end)
 				request_resource(res, &crashk_res);
--- a/arch/x86/kernel/efi_32.c
+++ b/arch/x86/kernel/efi_32.c
@@ -49,6 +49,9 @@ EXPORT_SYMBOL(efi);
 static struct efi efi_phys;
 struct efi_memory_map memmap;
 
+extern struct resource iomem_resource;
+extern struct resource code_resource, data_resource, bss_resource;
+
 /*
  * We require an early boot_ioremap mapping mechanism initially
  */
@@ -599,8 +602,7 @@ void __init efi_enter_virtual_mode(void)
 }
 
 void __init
-efi_initialize_iomem_resources(struct resource *code_resource,
-			       struct resource *data_resource)
+efi_initialize_iomem_resources(void)
 {
 	struct resource *res;
 	efi_memory_desc_t *md;
@@ -670,8 +672,9 @@ efi_initialize_iomem_resources(struct re
 		 * it repeatedly and let the resource manager test it.
 		 */
 		if (md->type == EFI_CONVENTIONAL_MEMORY) {
-			request_resource(res, code_resource);
-			request_resource(res, data_resource);
+			request_resource(res, &code_resource);
+			request_resource(res, &data_resource);
+			request_resource(res, &bss_resource);
 #ifdef CONFIG_KEXEC
 			request_resource(res, &crashk_res);
 #endif
--- a/arch/x86/kernel/setup_32.c
+++ b/arch/x86/kernel/setup_32.c
@@ -60,6 +60,7 @@
 #include <asm/vmi.h>
 #include <setup_arch.h>
 #include <bios_ebda.h>
+#include <asm/cacheflush.h>
 
 /* This value is set up by the early boot code to point to the value
    immediately after the boot time page tables.  It contains a *physical*
@@ -73,6 +74,7 @@ int disable_pse __devinitdata = 0;
  */
 extern struct resource code_resource;
 extern struct resource data_resource;
+extern struct resource bss_resource;
 
 /* cpu data as detected by the assembly code in head.S */
 struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
@@ -595,6 +597,8 @@ void __init setup_arch(char **cmdline_p)
 	code_resource.end = virt_to_phys(_etext)-1;
 	data_resource.start = virt_to_phys(_etext);
 	data_resource.end = virt_to_phys(_edata)-1;
+	bss_resource.start = virt_to_phys(&__bss_start);
+	bss_resource.end = virt_to_phys(&__bss_stop)-1;
 
 	parse_early_param();
 
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -59,6 +59,7 @@
 #include <asm/numa.h>
 #include <asm/sections.h>
 #include <asm/dmi.h>
+#include <asm/cacheflush.h>
 
 /*
  * Machine setup..
@@ -134,6 +135,12 @@ struct resource code_resource = {
 	.end = 0,
 	.flags = IORESOURCE_RAM,
 };
+struct resource bss_resource = {
+	.name = "Kernel bss",
+	.start = 0,
+	.end = 0,
+	.flags = IORESOURCE_RAM,
+};
 
 #ifdef CONFIG_PROC_VMCORE
 /* elfcorehdr= specifies the location of elf core header
@@ -276,6 +283,8 @@ void __init setup_arch(char **cmdline_p)
 	code_resource.end = virt_to_phys(&_etext)-1;
 	data_resource.start = virt_to_phys(&_etext);
 	data_resource.end = virt_to_phys(&_edata)-1;
+	bss_resource.start = virt_to_phys(&__bss_start);
+	bss_resource.end = virt_to_phys(&__bss_stop)-1;
 
 	early_identify_cpu(&boot_cpu_data);
 
--- a/include/linux/efi.h
+++ b/include/linux/efi.h
@@ -297,8 +297,7 @@ extern u64 efi_mem_attribute (unsigned l
 extern int efi_mem_attribute_range (unsigned long phys_addr, unsigned long size,
 				    u64 attr);
 extern int __init efi_uart_console_only (void);
-extern void efi_initialize_iomem_resources(struct resource *code_resource,
-					struct resource *data_resource);
+extern void efi_initialize_iomem_resources(void);
 extern unsigned long efi_get_time(void);
 extern int efi_set_rtc_mmss(unsigned long nowtime);
 extern int is_available_memory(efi_memory_desc_t * md);

-- 

[patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Bernhard Walle]

[-- Attachment #1: bootmem-alloc --]
[-- Type: text/plain, Size: 6344 bytes --]

This flag changes the reserve_bootmem() function to accept a new flag
BOOTMEM_EXCLUSIVE. If that flag is set, the function returns with
-EBUSY if the memory already has been reserved in the past. This is to
avoid conflicts.

IMPORTANT: The patch is only proof of concept. This means that it's only for
x86 and breaks other architectures. If the patch is ok, I'll change all other
architectures, too.


Signed-off-by: Bernhard Walle <bwalle@suse.de>

---
 arch/x86/kernel/mpparse_32.c |    4 ++--
 arch/x86/kernel/setup_32.c   |   12 ++++++------
 arch/x86/kernel/setup_64.c   |    2 +-
 include/linux/bootmem.h      |   13 ++++++++++++-
 mm/bootmem.c                 |   15 ++++++++++-----
 5 files changed, 31 insertions(+), 15 deletions(-)

--- a/arch/x86/kernel/mpparse_32.c
+++ b/arch/x86/kernel/mpparse_32.c
@@ -736,7 +736,7 @@ static int __init smp_scan_config (unsig
 			smp_found_config = 1;
 			printk(KERN_INFO "found SMP MP-table at %08lx\n",
 						virt_to_phys(mpf));
-			reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE);
+			reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE, 0);
 			if (mpf->mpf_physptr) {
 				/*
 				 * We cannot access to MPC table to compute
@@ -751,7 +751,7 @@ static int __init smp_scan_config (unsig
 				unsigned long end = max_low_pfn * PAGE_SIZE;
 				if (mpf->mpf_physptr + size > end)
 					size = end - mpf->mpf_physptr;
-				reserve_bootmem(mpf->mpf_physptr, size);
+				reserve_bootmem(mpf->mpf_physptr, size, 0);
 			}
 
 			mpf_found = mpf;
--- a/arch/x86/kernel/setup_32.c
+++ b/arch/x86/kernel/setup_32.c
@@ -317,7 +317,7 @@ static void __init reserve_ebda_region(v
 	unsigned int addr;
 	addr = get_bios_ebda();
 	if (addr)
-		reserve_bootmem(addr, PAGE_SIZE);	
+		reserve_bootmem(addr, PAGE_SIZE, 0);
 }
 
 #ifndef CONFIG_NEED_MULTIPLE_NODES
@@ -439,13 +439,13 @@ void __init setup_bootmem_allocator(void
 	 * bootmem allocator with an invalid RAM area.
 	 */
 	reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
-			 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text));
+			 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text), 0);
 
 	/*
 	 * reserve physical page 0 - it's a special BIOS page on many boxes,
 	 * enabling clean reboots, SMP operation, laptop functions.
 	 */
-	reserve_bootmem(0, PAGE_SIZE);
+	reserve_bootmem(0, PAGE_SIZE, 0);
 
 	/* reserve EBDA region, it's a 4K region */
 	reserve_ebda_region();
@@ -455,7 +455,7 @@ void __init setup_bootmem_allocator(void
        unless you have no PS/2 mouse plugged in. */
 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 	    boot_cpu_data.x86 == 6)
-	     reserve_bootmem(0xa0000 - 4096, 4096);
+	     reserve_bootmem(0xa0000 - 4096, 4096, 0);
 
 #ifdef CONFIG_SMP
 	/*
@@ -463,7 +463,7 @@ void __init setup_bootmem_allocator(void
 	 * FIXME: Don't need the extra page at 4K, but need to fix
 	 * trampoline before removing it. (see the GDT stuff)
 	 */
-	reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
+	reserve_bootmem(PAGE_SIZE, PAGE_SIZE, 0);
 #endif
 #ifdef CONFIG_ACPI_SLEEP
 	/*
@@ -481,7 +481,7 @@ void __init setup_bootmem_allocator(void
 #ifdef CONFIG_BLK_DEV_INITRD
 	if (LOADER_TYPE && INITRD_START) {
 		if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
-			reserve_bootmem(INITRD_START, INITRD_SIZE);
+			reserve_bootmem(INITRD_START, INITRD_SIZE, 0);
 			initrd_start = INITRD_START + PAGE_OFFSET;
 			initrd_end = initrd_start+INITRD_SIZE;
 		}
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -218,7 +218,7 @@ static void __init reserve_crashkernel(v
 					(unsigned long)(free_mem >> 20));
 			crashk_res.start = crash_base;
 			crashk_res.end   = crash_base + crash_size - 1;
-			reserve_bootmem(crash_base, crash_size);
+			reserve_bootmem(crash_base, crash_size, 0);
 		} else
 			printk(KERN_INFO "crashkernel reservation failed - "
 					"you have to specify a base address\n");
--- a/include/linux/bootmem.h
+++ b/include/linux/bootmem.h
@@ -61,8 +61,19 @@ extern void *__alloc_bootmem_core(struct
 				  unsigned long limit);
 extern void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size);
 
+/*
+ * flags for reserve_bootmem (also if CONFIG_HAVE_ARCH_BOOTMEM_NODE,
+ * the architecture-specific code should honor this)
+ */
+#define BOOTMEM_EXCLUSIVE	(1<<0)
+
 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
-extern void reserve_bootmem(unsigned long addr, unsigned long size);
+/*
+ * If flags is 0, then the return value is always 0 (success). If
+ * flags contains BOOTMEM_EXCLUSIVE, then -EBUSY is returned if the
+ * memory already was reserved.
+ */
+extern int reserve_bootmem(unsigned long addr, unsigned long size, int flags);
 #define alloc_bootmem(x) \
 	__alloc_bootmem(x, SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS))
 #define alloc_bootmem_low(x) \
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -111,8 +111,8 @@ static unsigned long __init init_bootmem
  * might be used for boot-time allocations - or it might get added
  * to the free page pool later on.
  */
-static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
-					unsigned long size)
+static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
+					unsigned long size, int flags)
 {
 	unsigned long sidx, eidx;
 	unsigned long i;
@@ -133,7 +133,11 @@ static void __init reserve_bootmem_core(
 #ifdef CONFIG_DEBUG_BOOTMEM
 			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 #endif
+			if (flags & BOOTMEM_EXCLUSIVE)
+				return -EBUSY;
 		}
+
+	return 0;
 }
 
 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
@@ -376,7 +380,7 @@ unsigned long __init init_bootmem_node(p
 void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 				 unsigned long size)
 {
-	reserve_bootmem_core(pgdat->bdata, physaddr, size);
+	reserve_bootmem_core(pgdat->bdata, physaddr, size, 0);
 }
 
 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
@@ -398,9 +402,10 @@ unsigned long __init init_bootmem(unsign
 }
 
 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
-void __init reserve_bootmem(unsigned long addr, unsigned long size)
+int __init reserve_bootmem(unsigned long addr, unsigned long size,
+			    int flags)
 {
-	reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
+	return reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size, flags);
 }
 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 

-- 

[patch 3/3] Use BOOTMEM_EXCLUSIVE on x86

[Bernhard Walle]

[-- Attachment #1: bootmem-alloc-use-crashkernel --]
[-- Type: text/plain, Size: 3454 bytes --]

This patch uses the BOOTMEM_EXCLUSIVE, introduced in the previous patch,
to avoid conflicts while reserving the memory for the kdump carpture kernel
(crashkernel=).

The modification has been tested on i386.


Signed-off-by: Bernhard Walle <bwalle@suse.de>

---
 arch/x86/kernel/setup_32.c |   28 ++++++++++++++++++----------
 arch/x86/kernel/setup_64.c |   34 +++++++++++++++++++++-------------
 2 files changed, 39 insertions(+), 23 deletions(-)

--- a/arch/x86/kernel/setup_32.c
+++ b/arch/x86/kernel/setup_32.c
@@ -403,18 +403,26 @@ static void __init reserve_crashkernel(v
 	ret = parse_crashkernel(boot_command_line, total_mem,
 			&crash_size, &crash_base);
 	if (ret == 0 && crash_size > 0) {
-		if (crash_base > 0) {
-			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
-					"for crashkernel (System RAM: %ldMB)\n",
-					(unsigned long)(crash_size >> 20),
-					(unsigned long)(crash_base >> 20),
-					(unsigned long)(total_mem >> 20));
-			crashk_res.start = crash_base;
-			crashk_res.end   = crash_base + crash_size - 1;
-			reserve_bootmem(crash_base, crash_size);
-		} else
+		if (crash_base <= 0) {
 			printk(KERN_INFO "crashkernel reservation failed - "
 					"you have to specify a base address\n");
+			return;
+		}
+
+		if (reserve_bootmem(crash_base, crash_size,
+					BOOTMEM_EXCLUSIVE) < 0) {
+			printk(KERN_INFO "crashkernel reservation failed - "
+					"memory is in use\n");
+			return;
+		}
+
+		printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
+				"for crashkernel (System RAM: %ldMB)\n",
+				(unsigned long)(crash_size >> 20),
+				(unsigned long)(crash_base >> 20),
+				(unsigned long)(total_mem >> 20));
+		crashk_res.start = crash_base;
+		crashk_res.end   = crash_base + crash_size - 1;
 	}
 }
 #else
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -201,27 +201,35 @@ static inline void copy_edd(void)
 #ifdef CONFIG_KEXEC
 static void __init reserve_crashkernel(void)
 {
-	unsigned long long free_mem;
+	unsigned long long total_mem;
 	unsigned long long crash_size, crash_base;
 	int ret;
 
-	free_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
+	total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
 
-	ret = parse_crashkernel(boot_command_line, free_mem,
+	ret = parse_crashkernel(boot_command_line, total_mem,
 			&crash_size, &crash_base);
 	if (ret == 0 && crash_size) {
-		if (crash_base > 0) {
-			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
-					"for crashkernel (System RAM: %ldMB)\n",
-					(unsigned long)(crash_size >> 20),
-					(unsigned long)(crash_base >> 20),
-					(unsigned long)(free_mem >> 20));
-			crashk_res.start = crash_base;
-			crashk_res.end   = crash_base + crash_size - 1;
-			reserve_bootmem(crash_base, crash_size, 0);
-		} else
+		if (crash_base <= 0) {
 			printk(KERN_INFO "crashkernel reservation failed - "
 					"you have to specify a base address\n");
+			return;
+		}
+
+		if (reserve_bootmem(crash_base, crash_size,
+					BOOTMEM_EXCLUSIVE) < 0) {
+			printk(KERN_INFO "crashkernel reservation failed - "
+					"memory is in use\n");
+			return;
+		}
+
+		printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
+				"for crashkernel (System RAM: %ldMB)\n",
+				(unsigned long)(crash_size >> 20),
+				(unsigned long)(crash_base >> 20),
+				(unsigned long)(total_mem >> 20));
+		crashk_res.start = crash_base;
+		crashk_res.end   = crash_base + crash_size - 1;
 	}
 }
 #else

-- 

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Dave Hansen]

On Tue, 2007-10-16 at 18:28 +0200, Bernhard Walle wrote:
> 
> @@ -736,7 +736,7 @@ static int __init smp_scan_config (unsig
>                         smp_found_config = 1;
>                         printk(KERN_INFO "found SMP MP-table at %08lx\n",
>                                                 virt_to_phys(mpf));
> -                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE);
> +                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE, 0);
>                         if (mpf->mpf_physptr) {
>                                 /* 

Could you give all of these 0's a name?  I really hate seeing random
magic numbers in these things.  0 completely kills the ability of
someone to read the code and figure out what it is trying to do without
going and looking at reserve_bootmem().

Or, alternatively, do something like this:

-extern void reserve_bootmem(unsigned long addr, unsigned long size);
+/*
+ * If flags is 0, then the return value is always 0 (success). If
+ * flags contains BOOTMEM_EXCLUSIVE, then -EBUSY is returned if the
+ * memory already was reserved.
+ */
+extern int reserve_bootmem(unsigned long addr, unsigned long size, int flag);
+int reserve_bootmem(unsigned long addr, unsigned long size)
+{
+	/* the 0 is because we don't
+	return reserve_bootmem_exclusive(addr, size, 0);
+}

Where all of the existing callers stay the same.  But, the ones wanting
exclusive access actually call the _exclusive() variant.  

-- Dave

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Bernhard Walle]

Hi,

* Dave Hansen <haveblue@us.ibm.com> [2007-10-16 20:08]:
> On Tue, 2007-10-16 at 18:28 +0200, Bernhard Walle wrote:
> > 
> > @@ -736,7 +736,7 @@ static int __init smp_scan_config (unsig
> >                         smp_found_config = 1;
> >                         printk(KERN_INFO "found SMP MP-table at %08lx\n",
> >                                                 virt_to_phys(mpf));
> > -                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE);
> > +                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE, 0);
> >                         if (mpf->mpf_physptr) {
> >                                 /* 
> 
> Could you give all of these 0's a name?  I really hate seeing random
> magic numbers in these things.  0 completely kills the ability of
> someone to read the code and figure out what it is trying to do without
> going and looking at reserve_bootmem().

Of course I can replace that zeroes with something like BOOTMEM_DEFAULT.

> Or, alternatively, do something like this:
> 
> -extern void reserve_bootmem(unsigned long addr, unsigned long size);

Andi was against more bootmem functions. ;)


Thanks,
   Bernhard

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Dave Hansen]

On Tue, 2007-10-16 at 20:44 +0200, Bernhard Walle wrote:
> * Dave Hansen <haveblue@us.ibm.com> [2007-10-16 20:08]:
> > On Tue, 2007-10-16 at 18:28 +0200, Bernhard Walle wrote:
> > > 
> > > @@ -736,7 +736,7 @@ static int __init smp_scan_config (unsig
> > >                         smp_found_config = 1;
> > >                         printk(KERN_INFO "found SMP MP-table at %08lx\n",
> > >                                                 virt_to_phys(mpf));
> > > -                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE);
> > > +                       reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE, 0);
> > >                         if (mpf->mpf_physptr) {
> > >                                 /* 
> > 
> > Could you give all of these 0's a name?  I really hate seeing random
> > magic numbers in these things.  0 completely kills the ability of
> > someone to read the code and figure out what it is trying to do without
> > going and looking at reserve_bootmem().
> 
> Of course I can replace that zeroes with something like BOOTMEM_DEFAULT.

Cool.

> > Or, alternatively, do something like this:
> > 
> > -extern void reserve_bootmem(unsigned long addr, unsigned long size);
> 
> Andi was against more bootmem functions. ;)

Yeah, I can't really blame him.

-- Dave

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Vivek Goyal]

[..]
> +/*
> + * If flags is 0, then the return value is always 0 (success). If
> + * flags contains BOOTMEM_EXCLUSIVE, then -EBUSY is returned if the
> + * memory already was reserved.
> + */
> +extern int reserve_bootmem(unsigned long addr, unsigned long size, int flags);
>  #define alloc_bootmem(x) \
>  	__alloc_bootmem(x, SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS))
>  #define alloc_bootmem_low(x) \
> --- a/mm/bootmem.c
> +++ b/mm/bootmem.c
> @@ -111,8 +111,8 @@ static unsigned long __init init_bootmem
>   * might be used for boot-time allocations - or it might get added
>   * to the free page pool later on.
>   */
> -static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> -					unsigned long size)
> +static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> +					unsigned long size, int flags)
>  {
>  	unsigned long sidx, eidx;
>  	unsigned long i;
> @@ -133,7 +133,11 @@ static void __init reserve_bootmem_core(
>  #ifdef CONFIG_DEBUG_BOOTMEM
>  			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
>  #endif
> +			if (flags & BOOTMEM_EXCLUSIVE)
> +				return -EBUSY;

I think we should unreserve the chunks of memory we have reserved so
far (Memory reserved from sidx to i), in case of error.

Thanks
Vivek

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Bernhard Walle]

* Vivek Goyal <vgoyal@in.ibm.com> [2007-10-17 13:05]:
> 
> [..]
> > +/*
> > + * If flags is 0, then the return value is always 0 (success). If
> > + * flags contains BOOTMEM_EXCLUSIVE, then -EBUSY is returned if the
> > + * memory already was reserved.
> > + */
> > +extern int reserve_bootmem(unsigned long addr, unsigned long size, int flags);
> >  #define alloc_bootmem(x) \
> >  	__alloc_bootmem(x, SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS))
> >  #define alloc_bootmem_low(x) \
> > --- a/mm/bootmem.c
> > +++ b/mm/bootmem.c
> > @@ -111,8 +111,8 @@ static unsigned long __init init_bootmem
> >   * might be used for boot-time allocations - or it might get added
> >   * to the free page pool later on.
> >   */
> > -static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> > -					unsigned long size)
> > +static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> > +					unsigned long size, int flags)
> >  {
> >  	unsigned long sidx, eidx;
> >  	unsigned long i;
> > @@ -133,7 +133,11 @@ static void __init reserve_bootmem_core(
> >  #ifdef CONFIG_DEBUG_BOOTMEM
> >  			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
> >  #endif
> > +			if (flags & BOOTMEM_EXCLUSIVE)
> > +				return -EBUSY;
> 
> I think we should unreserve the chunks of memory we have reserved so
> far (Memory reserved from sidx to i), in case of error.

Unfortunately, that's not possible without using a lock (or counters
instead of a bitmap) any more. If we just do

	for (i--; i >= sidx; i--)
		clear_bit(i, bdata->node_bootmem_map);

then another thread of execution could reserve the memory (without
BOOTMEM_EXCLUSIVE) in between -- and the code would free the memory
which is already reserved.

I think that could be modelled with a rwlock, not changing the default
case where BOOTMEM_EXCLUSIVE is not specified.


Thanks,
   Bernhard

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Vivek Goyal]

On Wed, Oct 17, 2007 at 01:36:51PM +0200, Bernhard Walle wrote:
[..]
> > > +static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
> > > +					unsigned long size, int flags)
> > >  {
> > >  	unsigned long sidx, eidx;
> > >  	unsigned long i;
> > > @@ -133,7 +133,11 @@ static void __init reserve_bootmem_core(
> > >  #ifdef CONFIG_DEBUG_BOOTMEM
> > >  			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
> > >  #endif
> > > +			if (flags & BOOTMEM_EXCLUSIVE)
> > > +				return -EBUSY;
> > 
> > I think we should unreserve the chunks of memory we have reserved so
> > far (Memory reserved from sidx to i), in case of error.
> 
> Unfortunately, that's not possible without using a lock (or counters
> instead of a bitmap) any more. If we just do
> 
> 	for (i--; i >= sidx; i--)
> 		clear_bit(i, bdata->node_bootmem_map);
> 
> then another thread of execution could reserve the memory (without
> BOOTMEM_EXCLUSIVE) in between -- and the code would free the memory
> which is already reserved.
> 
> I think that could be modelled with a rwlock, not changing the default
> case where BOOTMEM_EXCLUSIVE is not specified.

SMP initialization takes place after bootmem allocator has retired. That
would mean only one thread will be using bootmem allocator. Hence I think
unreserving memory without any kind of locking should be safe.

Thanks
Vivek

Re: [patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Bernhard Walle]

* Vivek Goyal <vgoyal@in.ibm.com> [2007-10-17 13:05]:
> 
> 
> I think we should unreserve the chunks of memory we have reserved so
> far (Memory reserved from sidx to i), in case of error.

True. Next version is coming.


Thanks,
   Bernhard

[patch 2/3] Introduce BOOTMEM_EXCLUSIVE

[Bernhard Walle]

[-- Attachment #1: bootmem-alloc --]
[-- Type: text/plain, Size: 7865 bytes --]

This flag changes the reserve_bootmem() function to accept a new flag
BOOTMEM_EXCLUSIVE. If that flag is set, the function returns with
-EBUSY if the memory already has been reserved in the past. This is to
avoid conflicts.

Because it's necessary to unreserve the bootmem if a collision is discovered in
the middle of the area, a rwlock is introduced: only one BOOTMEM_EXCLUSIVE
caller is possible, but multiple BOOTMEM_DEFAULT callers. But if a
BOOTMEM_EXCLUSIVE caller is in reserve_bootmem_core(), no BOOTMEM_DEFAULT
callers are allowd.

IMPORTANT: The patch is only proof of concept. This means that it's only for
x86 and breaks other architectures. If the patch is ok, I'll change all other
architectures, too.


Signed-off-by: Bernhard Walle <bwalle@suse.de>

---
 arch/x86/kernel/mpparse_32.c |    6 ++++--
 arch/x86/kernel/setup_32.c   |   15 ++++++++-------
 arch/x86/kernel/setup_64.c   |    5 +++--
 include/linux/bootmem.h      |   14 +++++++++++++-
 mm/bootmem.c                 |   25 ++++++++++++++++++++-----
 5 files changed, 48 insertions(+), 17 deletions(-)

--- a/arch/x86/kernel/mpparse_32.c
+++ b/arch/x86/kernel/mpparse_32.c
@@ -736,7 +736,8 @@ static int __init smp_scan_config (unsig
 			smp_found_config = 1;
 			printk(KERN_INFO "found SMP MP-table at %08lx\n",
 						virt_to_phys(mpf));
-			reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE);
+			reserve_bootmem(virt_to_phys(mpf), PAGE_SIZE,
+					BOOTMEM_DEFAULT);
 			if (mpf->mpf_physptr) {
 				/*
 				 * We cannot access to MPC table to compute
@@ -751,7 +752,8 @@ static int __init smp_scan_config (unsig
 				unsigned long end = max_low_pfn * PAGE_SIZE;
 				if (mpf->mpf_physptr + size > end)
 					size = end - mpf->mpf_physptr;
-				reserve_bootmem(mpf->mpf_physptr, size);
+				reserve_bootmem(mpf->mpf_physptr, size,
+						BOOTMEM_DEFAULT);
 			}
 
 			mpf_found = mpf;
--- a/arch/x86/kernel/setup_32.c
+++ b/arch/x86/kernel/setup_32.c
@@ -317,7 +317,7 @@ static void __init reserve_ebda_region(v
 	unsigned int addr;
 	addr = get_bios_ebda();
 	if (addr)
-		reserve_bootmem(addr, PAGE_SIZE);	
+		reserve_bootmem(addr, PAGE_SIZE, BOOTMEM_DEFAULT);
 }
 
 #ifndef CONFIG_NEED_MULTIPLE_NODES
@@ -411,7 +411,7 @@ static void __init reserve_crashkernel(v
 					(unsigned long)(total_mem >> 20));
 			crashk_res.start = crash_base;
 			crashk_res.end   = crash_base + crash_size - 1;
-			reserve_bootmem(crash_base, crash_size);
+			reserve_bootmem(crash_base, crash_size, BOOTMEM_DEFAULT);
 		} else
 			printk(KERN_INFO "crashkernel reservation failed - "
 					"you have to specify a base address\n");
@@ -439,13 +439,14 @@ void __init setup_bootmem_allocator(void
 	 * bootmem allocator with an invalid RAM area.
 	 */
 	reserve_bootmem(__pa_symbol(_text), (PFN_PHYS(min_low_pfn) +
-			 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text));
+			 bootmap_size + PAGE_SIZE-1) - __pa_symbol(_text),
+			 BOOTMEM_DEFAULT);
 
 	/*
 	 * reserve physical page 0 - it's a special BIOS page on many boxes,
 	 * enabling clean reboots, SMP operation, laptop functions.
 	 */
-	reserve_bootmem(0, PAGE_SIZE);
+	reserve_bootmem(0, PAGE_SIZE, BOOTMEM_DEFAULT);
 
 	/* reserve EBDA region, it's a 4K region */
 	reserve_ebda_region();
@@ -455,7 +456,7 @@ void __init setup_bootmem_allocator(void
        unless you have no PS/2 mouse plugged in. */
 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 	    boot_cpu_data.x86 == 6)
-	     reserve_bootmem(0xa0000 - 4096, 4096);
+	     reserve_bootmem(0xa0000 - 4096, 4096, BOOTMEM_DEFAULT);
 
 #ifdef CONFIG_SMP
 	/*
@@ -463,7 +464,7 @@ void __init setup_bootmem_allocator(void
 	 * FIXME: Don't need the extra page at 4K, but need to fix
 	 * trampoline before removing it. (see the GDT stuff)
 	 */
-	reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
+	reserve_bootmem(PAGE_SIZE, PAGE_SIZE, BOOTMEM_DEFAULT);
 #endif
 #ifdef CONFIG_ACPI_SLEEP
 	/*
@@ -481,7 +482,7 @@ void __init setup_bootmem_allocator(void
 #ifdef CONFIG_BLK_DEV_INITRD
 	if (LOADER_TYPE && INITRD_START) {
 		if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
-			reserve_bootmem(INITRD_START, INITRD_SIZE);
+			reserve_bootmem(INITRD_START, INITRD_SIZE, BOOTMEM_DEFAULT);
 			initrd_start = INITRD_START + PAGE_OFFSET;
 			initrd_end = initrd_start+INITRD_SIZE;
 		}
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -171,7 +171,7 @@ contig_initmem_init(unsigned long start_
 	bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
 	e820_register_active_regions(0, start_pfn, end_pfn);
 	free_bootmem_with_active_regions(0, end_pfn);
-	reserve_bootmem(bootmap, bootmap_size);
+	reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 } 
 #endif
 
@@ -218,7 +218,8 @@ static void __init reserve_crashkernel(v
 					(unsigned long)(free_mem >> 20));
 			crashk_res.start = crash_base;
 			crashk_res.end   = crash_base + crash_size - 1;
-			reserve_bootmem(crash_base, crash_size);
+			reserve_bootmem(crash_base, crash_size,
+					BOOTMEM_DEFAULT);
 		} else
 			printk(KERN_INFO "crashkernel reservation failed - "
 					"you have to specify a base address\n");
--- a/include/linux/bootmem.h
+++ b/include/linux/bootmem.h
@@ -61,8 +61,20 @@ extern void *__alloc_bootmem_core(struct
 				  unsigned long limit);
 extern void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size);
 
+/*
+ * flags for reserve_bootmem (also if CONFIG_HAVE_ARCH_BOOTMEM_NODE,
+ * the architecture-specific code should honor this)
+ */
+#define BOOTMEM_DEFAULT		0
+#define BOOTMEM_EXCLUSIVE	(1<<0)
+
 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
-extern void reserve_bootmem(unsigned long addr, unsigned long size);
+/*
+ * If flags is 0, then the return value is always 0 (success). If
+ * flags contains BOOTMEM_EXCLUSIVE, then -EBUSY is returned if the
+ * memory already was reserved.
+ */
+extern int reserve_bootmem(unsigned long addr, unsigned long size, int flags);
 #define alloc_bootmem(x) \
 	__alloc_bootmem(x, SMP_CACHE_BYTES, __pa(MAX_DMA_ADDRESS))
 #define alloc_bootmem_low(x) \
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -111,11 +111,12 @@ static unsigned long __init init_bootmem
  * might be used for boot-time allocations - or it might get added
  * to the free page pool later on.
  */
-static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
-					unsigned long size)
+static int __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
+					unsigned long size, int flags)
 {
 	unsigned long sidx, eidx;
 	unsigned long i;
+	int ret;
 
 	/*
 	 * round up, partially reserved pages are considered
@@ -133,7 +134,20 @@ static void __init reserve_bootmem_core(
 #ifdef CONFIG_DEBUG_BOOTMEM
 			printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 #endif
+			if (flags & BOOTMEM_EXCLUSIVE) {
+				ret = -EBUSY;
+				goto err;
+			}
 		}
+
+	return 0;
+
+err:
+	/* unreserve memory we accidentally reserved */
+	for (i--; i >= sidx; i--)
+		clear_bit(i, bdata->node_bootmem_map);
+
+	return ret;
 }
 
 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
@@ -376,7 +390,7 @@ unsigned long __init init_bootmem_node(p
 void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 				 unsigned long size)
 {
-	reserve_bootmem_core(pgdat->bdata, physaddr, size);
+	reserve_bootmem_core(pgdat->bdata, physaddr, size, BOOTMEM_DEFAULT);
 }
 
 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
@@ -398,9 +412,10 @@ unsigned long __init init_bootmem(unsign
 }
 
 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
-void __init reserve_bootmem(unsigned long addr, unsigned long size)
+int __init reserve_bootmem(unsigned long addr, unsigned long size,
+			    int flags)
 {
-	reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
+	return reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size, flags);
 }
 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 

--