| Message ID | 1470681607-36883-2-git-send-email-thgarnie@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Aug 08, 2016 at 11:40:07AM -0700, Thomas Garnier wrote: > Default implementation expects 6 pages maximum are needed for low page > allocations. If KASLR memory randomization is enabled, the worse case > of e820 layout would require 12 pages (no large pages). It is due to the > PUD level randomization and the variable e820 memory layout. Can't this number of required pages be computed based on the randomization offset or somesuch instead of maxing it out by default on every machine which enables CONFIG_RANDOMIZE_MEMORY? > This bug was found while doing extensive testing of KASLR memory > randomization on different type of hardware. > > Signed-off-by: Thomas Garnier <thgarnie@google.com> > --- > Based on next-20160805 > --- > arch/x86/mm/init.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c > index 6209289..3a27e6a 100644 > --- a/arch/x86/mm/init.c > +++ b/arch/x86/mm/init.c > @@ -130,6 +130,14 @@ void __init early_alloc_pgt_buf(void) > unsigned long tables = INIT_PGT_BUF_SIZE; > phys_addr_t base; > > + /* > + * Depending on the machine e860 memory layout and the PUD alignement. e820 alignment > + * We may need twice more pages when KASLR memoy randomization is memory > + * enabled. > + */ > + if (IS_ENABLED(CONFIG_RANDOMIZE_MEMORY)) > + tables *= 2; > + > base = __pa(extend_brk(tables, PAGE_SIZE)); > > pgt_buf_start = base >> PAGE_SHIFT; > -- > 2.8.0.rc3.226.g39d4020 >
On August 8, 2016 9:01:42 PM PDT, Borislav Petkov <bp@suse.de> wrote: >On Mon, Aug 08, 2016 at 11:40:07AM -0700, Thomas Garnier wrote: >> Default implementation expects 6 pages maximum are needed for low >page >> allocations. If KASLR memory randomization is enabled, the worse case >> of e820 layout would require 12 pages (no large pages). It is due to >the >> PUD level randomization and the variable e820 memory layout. > >Can't this number of required pages be computed based on the >randomization offset or somesuch instead of maxing it out by default on >every machine which enables CONFIG_RANDOMIZE_MEMORY? > >> This bug was found while doing extensive testing of KASLR memory >> randomization on different type of hardware. >> >> Signed-off-by: Thomas Garnier <thgarnie@google.com> >> --- >> Based on next-20160805 >> --- >> arch/x86/mm/init.c | 8 ++++++++ >> 1 file changed, 8 insertions(+) >> >> diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c >> index 6209289..3a27e6a 100644 >> --- a/arch/x86/mm/init.c >> +++ b/arch/x86/mm/init.c >> @@ -130,6 +130,14 @@ void __init early_alloc_pgt_buf(void) >> unsigned long tables = INIT_PGT_BUF_SIZE; >> phys_addr_t base; >> >> + /* >> + * Depending on the machine e860 memory layout and the PUD >alignement. > > e820 alignment > >> + * We may need twice more pages when KASLR memoy randomization is > > memory > >> + * enabled. >> + */ >> + if (IS_ENABLED(CONFIG_RANDOMIZE_MEMORY)) >> + tables *= 2; >> + >> base = __pa(extend_brk(tables, PAGE_SIZE)); >> >> pgt_buf_start = base >> PAGE_SHIFT; >> -- >> 2.8.0.rc3.226.g39d4020 >> Upping the brk by a few pages is not a big deal. Unused brk gets reclaimed at init release time.
On 08/08/2016 09:40 PM, Thomas Garnier wrote: > Default implementation expects 6 pages maximum are needed for low page > allocations. If KASLR memory randomization is enabled, the worse case > of e820 layout would require 12 pages (no large pages). It is due to the > PUD level randomization and the variable e820 memory layout. > > This bug was found while doing extensive testing of KASLR memory > randomization on different type of hardware. > > Signed-off-by: Thomas Garnier <thgarnie@google.com> > --- > Based on next-20160805 > --- > arch/x86/mm/init.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c > index 6209289..3a27e6a 100644 > --- a/arch/x86/mm/init.c > +++ b/arch/x86/mm/init.c > @@ -130,6 +130,14 @@ void __init early_alloc_pgt_buf(void) > unsigned long tables = INIT_PGT_BUF_SIZE; > phys_addr_t base; > > + /* > + * Depending on the machine e860 memory layout and the PUD alignement. > + * We may need twice more pages when KASLR memoy randomization is > + * enabled. > + */ > + if (IS_ENABLED(CONFIG_RANDOMIZE_MEMORY)) > + tables *= 2; > + > base = __pa(extend_brk(tables, PAGE_SIZE)); > > pgt_buf_start = base >> PAGE_SHIFT; > You should increase the reserve also : RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); --Mika
On Mon, Aug 8, 2016 at 10:16 PM, Mika Penttilä <mika.penttila@nextfour.com> wrote: > On 08/08/2016 09:40 PM, Thomas Garnier wrote: >> Default implementation expects 6 pages maximum are needed for low page >> allocations. If KASLR memory randomization is enabled, the worse case >> of e820 layout would require 12 pages (no large pages). It is due to the >> PUD level randomization and the variable e820 memory layout. >> >> This bug was found while doing extensive testing of KASLR memory >> randomization on different type of hardware. >> >> Signed-off-by: Thomas Garnier <thgarnie@google.com> >> --- >> Based on next-20160805 >> --- >> arch/x86/mm/init.c | 8 ++++++++ >> 1 file changed, 8 insertions(+) >> >> diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c >> index 6209289..3a27e6a 100644 >> --- a/arch/x86/mm/init.c >> +++ b/arch/x86/mm/init.c >> @@ -130,6 +130,14 @@ void __init early_alloc_pgt_buf(void) >> unsigned long tables = INIT_PGT_BUF_SIZE; >> phys_addr_t base; >> >> + /* >> + * Depending on the machine e860 memory layout and the PUD alignement. >> + * We may need twice more pages when KASLR memoy randomization is >> + * enabled. >> + */ >> + if (IS_ENABLED(CONFIG_RANDOMIZE_MEMORY)) >> + tables *= 2; >> + >> base = __pa(extend_brk(tables, PAGE_SIZE)); >> >> pgt_buf_start = base >> PAGE_SHIFT; >> > > You should increase the reserve also : > RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); Yes, that's right. I will resend v2 changing INIT_PGD_BUF_SIZE. > > > --Mika >
diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c index 6209289..3a27e6a 100644 --- a/arch/x86/mm/init.c +++ b/arch/x86/mm/init.c @@ -130,6 +130,14 @@ void __init early_alloc_pgt_buf(void) unsigned long tables = INIT_PGT_BUF_SIZE; phys_addr_t base; + /* + * Depending on the machine e860 memory layout and the PUD alignement. + * We may need twice more pages when KASLR memoy randomization is + * enabled. + */ + if (IS_ENABLED(CONFIG_RANDOMIZE_MEMORY)) + tables *= 2; + base = __pa(extend_brk(tables, PAGE_SIZE)); pgt_buf_start = base >> PAGE_SHIFT;
Default implementation expects 6 pages maximum are needed for low page allocations. If KASLR memory randomization is enabled, the worse case of e820 layout would require 12 pages (no large pages). It is due to the PUD level randomization and the variable e820 memory layout. This bug was found while doing extensive testing of KASLR memory randomization on different type of hardware. Signed-off-by: Thomas Garnier <thgarnie@google.com> --- Based on next-20160805 --- arch/x86/mm/init.c | 8 ++++++++ 1 file changed, 8 insertions(+)