Message ID | 1473626125-13683-16-git-send-email-konrad.wilk@oracle.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
>>> On 11.09.16 at 22:35, <konrad.wilk@oracle.com> wrote: > Furthermore on x86 the bloat-o-meter detects that with this > change: > > [konrad@char xen]$ ~/linux/scripts/bloat-o-meter xen-syms xen-syms.align4 > add/remove: 0/0 grow/shrink: 3/14 up/down: 115/-1497 (-1382) > function old new delta > mod_l1_entry 1490 1587 +97 > p2m_switch_domain_altp2m_by_id 520 533 +13 > p2m_switch_vcpu_altp2m_by_id 453 458 +5 > machine_kexec 263 261 -2 > hvm_do_IRQ_dpci 244 242 -2 > sh_page_fault__guest_4 8252 8192 -60 > sh_audit_gw 1529 1462 -67 > validate_gl3e 337 264 -73 > validate_gl4e 449 375 -74 > p2m_altp2m_lazy_copy 730 652 -78 > set_typed_p2m_entry 1346 1259 -87 > virt_to_xen_l2e 491 365 -126 > sh_x86_emulate_write__guest_4 443 288 -155 > p2m_mem_access_check 1733 1576 -157 > sh_x86_emulate_cmpxchg__guest_4 512 349 -163 > __get_gfn_type_access 709 542 -167 > map_pages_to_xen 4430 4144 -286 > Total: Before=1974033, After=1972651, chg -0.07% > > We end up making the binary file a bit smaller. I'm fine with the change, but I'm having a very hard time buying the above: The change you do is to code used only in assembly files _and_ is not affecting .text (and I assume the sizes above are .text ones), yet all the functions listed above live in C ones. I can't help thinking that there must be something else going on, or that we had an actual problem before this. > --- a/xen/include/asm-x86/bug.h > +++ b/xen/include/asm-x86/bug.h > @@ -98,6 +98,7 @@ extern const struct bug_frame __start_bug_frames[], > .popsection > > .pushsection .bug_frames.\type, "a", @progbits > + .p2align 2 > .L\@bf: > .long (.L\@ud - .L\@bf) + \ > ((\line >> BUG_LINE_LO_WIDTH) << BUG_DISP_WIDTH) This ought to be accompanied by removing the enforcing of alignment in xen.lds.S. Jan
On Tue, Sep 13, 2016 at 03:21:04AM -0600, Jan Beulich wrote: > >>> On 11.09.16 at 22:35, <konrad.wilk@oracle.com> wrote: > > Furthermore on x86 the bloat-o-meter detects that with this > > change: > > > > [konrad@char xen]$ ~/linux/scripts/bloat-o-meter xen-syms xen-syms.align4 > > add/remove: 0/0 grow/shrink: 3/14 up/down: 115/-1497 (-1382) > > function old new delta > > mod_l1_entry 1490 1587 +97 > > p2m_switch_domain_altp2m_by_id 520 533 +13 > > p2m_switch_vcpu_altp2m_by_id 453 458 +5 > > machine_kexec 263 261 -2 > > hvm_do_IRQ_dpci 244 242 -2 > > sh_page_fault__guest_4 8252 8192 -60 > > sh_audit_gw 1529 1462 -67 > > validate_gl3e 337 264 -73 > > validate_gl4e 449 375 -74 > > p2m_altp2m_lazy_copy 730 652 -78 > > set_typed_p2m_entry 1346 1259 -87 > > virt_to_xen_l2e 491 365 -126 > > sh_x86_emulate_write__guest_4 443 288 -155 > > p2m_mem_access_check 1733 1576 -157 > > sh_x86_emulate_cmpxchg__guest_4 512 349 -163 > > __get_gfn_type_access 709 542 -167 > > map_pages_to_xen 4430 4144 -286 > > Total: Before=1974033, After=1972651, chg -0.07% > > > > We end up making the binary file a bit smaller. > > I'm fine with the change, but I'm having a very hard time buying the > above: The change you do is to code used only in assembly files _and_ That is indeed fishy. I re-ran the test and got: ~/linux/scripts/bloat-o-meter /tmp/xen-syms.orig /tmp/xen-syms.p2malign add/remove: 0/0 grow/shrink: 0/0 up/down: 0/0 (0) function old new delta which is more inline with what we expect. > is not affecting .text (and I assume the sizes above are .text ones), > yet all the functions listed above live in C ones. I can't help thinking > that there must be something else going on, or that we had an actual > problem before this. I am not sure what I managed to screw up. > > > --- a/xen/include/asm-x86/bug.h > > +++ b/xen/include/asm-x86/bug.h > > @@ -98,6 +98,7 @@ extern const struct bug_frame __start_bug_frames[], > > .popsection > > > > .pushsection .bug_frames.\type, "a", @progbits > > + .p2align 2 > > .L\@bf: > > .long (.L\@ud - .L\@bf) + \ > > ((\line >> BUG_LINE_LO_WIDTH) << BUG_DISP_WIDTH) > > This ought to be accompanied by removing the enforcing of alignment > in xen.lds.S. /me nods. Done! > > Jan >
diff --git a/xen/include/asm-arm/bug.h b/xen/include/asm-arm/bug.h index 68353e1..773d63e 100644 --- a/xen/include/asm-arm/bug.h +++ b/xen/include/asm-arm/bug.h @@ -52,6 +52,7 @@ struct bug_frame { ".popsection\n" \ ".pushsection .bug_frames." __stringify(type) ", \"a\", %progbits\n"\ "4:\n" \ + ".align 4\n" \ ".long (1b - 4b)\n" \ ".long (2b - 4b)\n" \ ".long (3b - 4b)\n" \ diff --git a/xen/include/asm-x86/bug.h b/xen/include/asm-x86/bug.h index c5d2d4c..9bb4a19 100644 --- a/xen/include/asm-x86/bug.h +++ b/xen/include/asm-x86/bug.h @@ -98,6 +98,7 @@ extern const struct bug_frame __start_bug_frames[], .popsection .pushsection .bug_frames.\type, "a", @progbits + .p2align 2 .L\@bf: .long (.L\@ud - .L\@bf) + \ ((\line >> BUG_LINE_LO_WIDTH) << BUG_DISP_WIDTH)
Most of the WARN_ON or BUG_ON sections are properly aligned on x86. However on ARM and on x86 assembler the macros don't include any aligment information - hence they end up being the default byte granularity. On ARM32 it is paramount that the aligment is word-size (4) otherwise if one tries to use (uint32_t*) access (such as livepatch ELF relocations) we get a Data Abort. Enforcing bug_frames to have the proper aligment across all architectures and in both C and x86 makes them all the same. Furthermore on x86 the bloat-o-meter detects that with this change: [konrad@char xen]$ ~/linux/scripts/bloat-o-meter xen-syms xen-syms.align4 add/remove: 0/0 grow/shrink: 3/14 up/down: 115/-1497 (-1382) function old new delta mod_l1_entry 1490 1587 +97 p2m_switch_domain_altp2m_by_id 520 533 +13 p2m_switch_vcpu_altp2m_by_id 453 458 +5 machine_kexec 263 261 -2 hvm_do_IRQ_dpci 244 242 -2 sh_page_fault__guest_4 8252 8192 -60 sh_audit_gw 1529 1462 -67 validate_gl3e 337 264 -73 validate_gl4e 449 375 -74 p2m_altp2m_lazy_copy 730 652 -78 set_typed_p2m_entry 1346 1259 -87 virt_to_xen_l2e 491 365 -126 sh_x86_emulate_write__guest_4 443 288 -155 p2m_mem_access_check 1733 1576 -157 sh_x86_emulate_cmpxchg__guest_4 512 349 -163 __get_gfn_type_access 709 542 -167 map_pages_to_xen 4430 4144 -286 Total: Before=1974033, After=1972651, chg -0.07% We end up making the binary file a bit smaller. On ARM32: add/remove: 1/0 grow/shrink: 0/1 up/down: 384/-288 (96) function old new delta gnttab_unpopulate_status_frames - 384 +384 do_grant_table_op 10808 10520 -288 And ARM64: add/remove: 1/2 grow/shrink: 0/1 up/down: 4164/-4236 (-72) function old new delta gnttab_map_grant_ref - 4164 +4164 do_grant_table_op 9892 9836 -56 grant_map_exists 300 - -300 __gnttab_map_grant_ref 3880 - -3880 Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> --- Cc: Julien Grall <julien.grall@arm.com> Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Jan Beulich <jbeulich@suse.com> Cc: Andrew Cooper <andrew.cooper3@citrix.com> v3: First submission. Replaces the "livepatch/elf: Adjust section aligment to word" patch. --- xen/include/asm-arm/bug.h | 1 + xen/include/asm-x86/bug.h | 1 + 2 files changed, 2 insertions(+)