| Message ID | 20240522131703.30839-3-andrew.cooper3@citrix.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | x86/shadow: Trace fixes and cleanup | expand |
On 22.05.2024 15:17, Andrew Cooper wrote: > trace_shadow_fixup() and trace_not_shadow_fault() both write out identical > trace records. Reimplement them in terms of a common sh_trace_gl1e_va(). > > There's no need to pack the trace record, even in the case of PAE paging. Isn't this altering the generated trace record for the 4-level case, in size changing from 20 to 24 bytes? > --- a/xen/arch/x86/mm/shadow/multi.c > +++ b/xen/arch/x86/mm/shadow/multi.c > @@ -1987,51 +1987,26 @@ static void sh_trace_va(uint32_t event, guest_va_t va) > sh_trace(event, sizeof(va), &va); > } > > -static inline void trace_shadow_fixup(guest_l1e_t gl1e, > - guest_va_t va) > +/* Shadow trace event with a gl1e, linear address and flags. */ > +static void sh_trace_gl1e_va(uint32_t event, guest_l1e_t gl1e, guest_va_t va) > { > if ( tb_init_done ) > { > - struct __packed { > - /* for PAE, guest_l1e may be 64 while guest_va may be 32; > - so put it first for alignment sake. */ > - guest_l1e_t gl1e; > - guest_va_t va; > - u32 flags; > - } d; > - u32 event; > - > - event = TRC_SHADOW_FIXUP | ((GUEST_PAGING_LEVELS-2)<<8); > - > - d.gl1e = gl1e; > - d.va = va; > - d.flags = this_cpu(trace_shadow_path_flags); > - > - trace(event, sizeof(d), &d); > - } > -} > - > -static inline void trace_not_shadow_fault(guest_l1e_t gl1e, > - guest_va_t va) > -{ > - if ( tb_init_done ) > - { > - struct __packed { > - /* for PAE, guest_l1e may be 64 while guest_va may be 32; > - so put it first for alignment sake. */ > + struct { > + /* > + * For GUEST_PAGING_LEVELS=3 (PAE paging), guest_l1e is 64 while > + * guest_va is 32. Put it first to avoid padding. > + */ > guest_l1e_t gl1e; > guest_va_t va; > - u32 flags; > - } d; > - u32 event; > - > - event = TRC_SHADOW_NOT_SHADOW | ((GUEST_PAGING_LEVELS-2)<<8); > - > - d.gl1e = gl1e; > - d.va = va; > - d.flags = this_cpu(trace_shadow_path_flags); > - > - trace(event, sizeof(d), &d); > + uint32_t flags; > + } d = { > + .gl1e = gl1e, > + .va = va, > + .flags = this_cpu(trace_shadow_path_flags), > + }; > + > + sh_trace(event, sizeof(d), &d); > } > } Unlike in patch 1, it's less clear here whether leaving the tb_init_done check is actually better to keep where it is. In principle the compiler should be able to re-arrange code enough to make it identical no matter which way it's written, at which point it might again be more desirable to have the check solely in sh_trace(). Jan
On 22/05/2024 2:47 pm, Jan Beulich wrote: > On 22.05.2024 15:17, Andrew Cooper wrote: >> trace_shadow_fixup() and trace_not_shadow_fault() both write out identical >> trace records. Reimplement them in terms of a common sh_trace_gl1e_va(). >> >> There's no need to pack the trace record, even in the case of PAE paging. > Isn't this altering the generated trace record for the 4-level case, in > size changing from 20 to 24 bytes? Oh, eww. Yes it does. I'll need to rework with __packed still in place. ~Andrew
diff --git a/xen/arch/x86/mm/shadow/multi.c b/xen/arch/x86/mm/shadow/multi.c index 1775952d7e18..f0a9cc527c0b 100644 --- a/xen/arch/x86/mm/shadow/multi.c +++ b/xen/arch/x86/mm/shadow/multi.c @@ -1987,51 +1987,26 @@ static void sh_trace_va(uint32_t event, guest_va_t va) sh_trace(event, sizeof(va), &va); } -static inline void trace_shadow_fixup(guest_l1e_t gl1e, - guest_va_t va) +/* Shadow trace event with a gl1e, linear address and flags. */ +static void sh_trace_gl1e_va(uint32_t event, guest_l1e_t gl1e, guest_va_t va) { if ( tb_init_done ) { - struct __packed { - /* for PAE, guest_l1e may be 64 while guest_va may be 32; - so put it first for alignment sake. */ - guest_l1e_t gl1e; - guest_va_t va; - u32 flags; - } d; - u32 event; - - event = TRC_SHADOW_FIXUP | ((GUEST_PAGING_LEVELS-2)<<8); - - d.gl1e = gl1e; - d.va = va; - d.flags = this_cpu(trace_shadow_path_flags); - - trace(event, sizeof(d), &d); - } -} - -static inline void trace_not_shadow_fault(guest_l1e_t gl1e, - guest_va_t va) -{ - if ( tb_init_done ) - { - struct __packed { - /* for PAE, guest_l1e may be 64 while guest_va may be 32; - so put it first for alignment sake. */ + struct { + /* + * For GUEST_PAGING_LEVELS=3 (PAE paging), guest_l1e is 64 while + * guest_va is 32. Put it first to avoid padding. + */ guest_l1e_t gl1e; guest_va_t va; - u32 flags; - } d; - u32 event; - - event = TRC_SHADOW_NOT_SHADOW | ((GUEST_PAGING_LEVELS-2)<<8); - - d.gl1e = gl1e; - d.va = va; - d.flags = this_cpu(trace_shadow_path_flags); - - trace(event, sizeof(d), &d); + uint32_t flags; + } d = { + .gl1e = gl1e, + .va = va, + .flags = this_cpu(trace_shadow_path_flags), + }; + + sh_trace(event, sizeof(d), &d); } } @@ -2603,7 +2578,7 @@ static int cf_check sh_page_fault( d->arch.paging.log_dirty.fault_count++; sh_reset_early_unshadow(v); - trace_shadow_fixup(gw.l1e, va); + sh_trace_gl1e_va(TRC_SHADOW_FIXUP, gw.l1e, va); done: __maybe_unused; sh_audit_gw(v, &gw); SHADOW_PRINTK("fixed\n"); @@ -2857,7 +2832,7 @@ static int cf_check sh_page_fault( put_gfn(d, gfn_x(gfn)); propagate: - trace_not_shadow_fault(gw.l1e, va); + sh_trace_gl1e_va(TRC_SHADOW_NOT_SHADOW, gw.l1e, va); return 0; }
trace_shadow_fixup() and trace_not_shadow_fault() both write out identical trace records. Reimplement them in terms of a common sh_trace_gl1e_va(). There's no need to pack the trace record, even in the case of PAE paging. No functional change. Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> --- CC: Jan Beulich <JBeulich@suse.com> CC: Roger Pau Monné <roger.pau@citrix.com> CC: George Dunlap <george.dunlap@citrix.com> v2: * New --- xen/arch/x86/mm/shadow/multi.c | 59 ++++++++++------------------------ 1 file changed, 17 insertions(+), 42 deletions(-)