| Message ID | 1467791668-2937838-1-git-send-email-snarpix@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 06/07/16 10:54, Stanislav Shmarov wrote: > In user-mode emulation Translation Block can consist of 2 guest pages. > In that case QEMU also mprotects 2 host pages that are dedicated for > guest memory, containing instructions. QEMU detects self-modifying code > with SEGFAULT signal processing. > > In case if instruction in 1st page is modifying memory of 2nd > page (or vice versa) QEMU will mark 2nd page with PAGE_WRITE, > invalidate TB, generate new TB contatining 1 guest instruction and > exit to CPU loop. QEMU won't call mprotect, and new TB will cause > same SEGFAULT. Page will have both PAGE_WRITE_ORG and PAGE_WRITE > flags, so QEMU will handle the signal as guest binary problem, > and exit with guest SEGFAULT. > > Solution is retranslate TB before marking pages as PAGE_WRITE, > and remove protection with mprotect on second SEGFAULT. > > Signed-off-by: Stanislav Shmarov <snarpix@gmail.com> Reviewed-by: Sergey Fedorov <sergey.fedorov@linaro.org> > --- > v2: Moved setting PAGE_WRITE flag to separte loop, to cover cases, > pointed by Sergey Fedorov. > > translate-all.c | 17 +++++++++++------ > 1 file changed, 11 insertions(+), 6 deletions(-) > > diff --git a/translate-all.c b/translate-all.c > index eaa95e4..fb3743f 100644 > --- a/translate-all.c > +++ b/translate-all.c > @@ -2020,13 +2020,8 @@ int page_unprotect(target_ulong address, uintptr_t pc) > host_start = address & qemu_host_page_mask; > host_end = host_start + qemu_host_page_size; > > - prot = 0; > for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) { > - p = page_find(addr >> TARGET_PAGE_BITS); > - p->flags |= PAGE_WRITE; > - prot |= p->flags; > - > - /* and since the content will be modified, we must invalidate > + /* Since the content will be modified, we must invalidate > the corresponding translated code. */ > if (tb_invalidate_phys_page(addr, pc)) { > mmap_unlock(); > @@ -2036,6 +2031,16 @@ int page_unprotect(target_ulong address, uintptr_t pc) > tb_invalidate_check(addr); > #endif > } > + > + /* If we got here, current TB have been retranslated (in case of > + * self-modifying code), now it's safe to remove page protection. > + */ > + prot = 0; > + for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) { > + p = page_find(addr >> TARGET_PAGE_BITS); > + p->flags |= PAGE_WRITE; > + prot |= p->flags; > + } > mprotect((void *)g2h(host_start), qemu_host_page_size, > prot & PAGE_BITS); >
May be this can be done faster. After invalidation of current TB, we know next executed TB will be 1 instruction long and will write to same address, and will cause segfault. In sigfault handler write protection to this page will be disabled. So, we can retranslate TB, and remove protection before execution of new TB. It will be one signal less. 2016-07-06 13:21 GMT+03:00 Sergey Fedorov <sergey.fedorov@linaro.org>: > On 06/07/16 10:54, Stanislav Shmarov wrote: > > In user-mode emulation Translation Block can consist of 2 guest pages. > > In that case QEMU also mprotects 2 host pages that are dedicated for > > guest memory, containing instructions. QEMU detects self-modifying code > > with SEGFAULT signal processing. > > > > In case if instruction in 1st page is modifying memory of 2nd > > page (or vice versa) QEMU will mark 2nd page with PAGE_WRITE, > > invalidate TB, generate new TB contatining 1 guest instruction and > > exit to CPU loop. QEMU won't call mprotect, and new TB will cause > > same SEGFAULT. Page will have both PAGE_WRITE_ORG and PAGE_WRITE > > flags, so QEMU will handle the signal as guest binary problem, > > and exit with guest SEGFAULT. > > > > Solution is retranslate TB before marking pages as PAGE_WRITE, > > and remove protection with mprotect on second SEGFAULT. > > > > Signed-off-by: Stanislav Shmarov <snarpix@gmail.com> > > Reviewed-by: Sergey Fedorov <sergey.fedorov@linaro.org> > > > --- > > v2: Moved setting PAGE_WRITE flag to separte loop, to cover cases, > > pointed by Sergey Fedorov. > > > > translate-all.c | 17 +++++++++++------ > > 1 file changed, 11 insertions(+), 6 deletions(-) > > > > diff --git a/translate-all.c b/translate-all.c > > index eaa95e4..fb3743f 100644 > > --- a/translate-all.c > > +++ b/translate-all.c > > @@ -2020,13 +2020,8 @@ int page_unprotect(target_ulong address, > uintptr_t pc) > > host_start = address & qemu_host_page_mask; > > host_end = host_start + qemu_host_page_size; > > > > - prot = 0; > > for (addr = host_start ; addr < host_end ; addr += > TARGET_PAGE_SIZE) { > > - p = page_find(addr >> TARGET_PAGE_BITS); > > - p->flags |= PAGE_WRITE; > > - prot |= p->flags; > > - > > - /* and since the content will be modified, we must > invalidate > > + /* Since the content will be modified, we must invalidate > > the corresponding translated code. */ > > if (tb_invalidate_phys_page(addr, pc)) { > > mmap_unlock(); > > @@ -2036,6 +2031,16 @@ int page_unprotect(target_ulong address, > uintptr_t pc) > > tb_invalidate_check(addr); > > #endif > > } > > + > > + /* If we got here, current TB have been retranslated (in case of > > + * self-modifying code), now it's safe to remove page > protection. > > + */ > > + prot = 0; > > + for (addr = host_start ; addr < host_end ; addr += > TARGET_PAGE_SIZE) { > > + p = page_find(addr >> TARGET_PAGE_BITS); > > + p->flags |= PAGE_WRITE; > > + prot |= p->flags; > > + } > > mprotect((void *)g2h(host_start), qemu_host_page_size, > > prot & PAGE_BITS); > > > >
On 06/07/16 15:01, Стас Шмаров wrote: > May be this can be done faster. > After invalidation of current TB, we know next executed TB will be 1 > instruction long and will write to same address, and will cause > segfault. In sigfault handler write protection to this page will be > disabled. > So, we can retranslate TB, and remove protection before execution of > new TB. It will be one signal less. You mean not to return 2 immediately but rather to remember the desired return value and keep invalidating guest pages, then go ahead and call mprotect() to enable writes to the host page? Thanks, Sergey > > 2016-07-06 13:21 GMT+03:00 Sergey Fedorov <sergey.fedorov@linaro.org > <mailto:sergey.fedorov@linaro.org>>: > > On 06/07/16 10:54, Stanislav Shmarov wrote: > > In user-mode emulation Translation Block can consist of 2 guest > pages. > > In that case QEMU also mprotects 2 host pages that are dedicated for > > guest memory, containing instructions. QEMU detects > self-modifying code > > with SEGFAULT signal processing. > > > > In case if instruction in 1st page is modifying memory of 2nd > > page (or vice versa) QEMU will mark 2nd page with PAGE_WRITE, > > invalidate TB, generate new TB contatining 1 guest instruction and > > exit to CPU loop. QEMU won't call mprotect, and new TB will cause > > same SEGFAULT. Page will have both PAGE_WRITE_ORG and PAGE_WRITE > > flags, so QEMU will handle the signal as guest binary problem, > > and exit with guest SEGFAULT. > > > > Solution is retranslate TB before marking pages as PAGE_WRITE, > > and remove protection with mprotect on second SEGFAULT. > > > > Signed-off-by: Stanislav Shmarov <snarpix@gmail.com > <mailto:snarpix@gmail.com>> > > Reviewed-by: Sergey Fedorov <sergey.fedorov@linaro.org > <mailto:sergey.fedorov@linaro.org>> > > > --- > > v2: Moved setting PAGE_WRITE flag to separte loop, to cover cases, > > pointed by Sergey Fedorov. > > > > translate-all.c | 17 +++++++++++------ > > 1 file changed, 11 insertions(+), 6 deletions(-) > > > > diff --git a/translate-all.c b/translate-all.c > > index eaa95e4..fb3743f 100644 > > --- a/translate-all.c > > +++ b/translate-all.c > > @@ -2020,13 +2020,8 @@ int page_unprotect(target_ulong address, > uintptr_t pc) > > host_start = address & qemu_host_page_mask; > > host_end = host_start + qemu_host_page_size; > > > > - prot = 0; > > for (addr = host_start ; addr < host_end ; addr += > TARGET_PAGE_SIZE) { > > - p = page_find(addr >> TARGET_PAGE_BITS); > > - p->flags |= PAGE_WRITE; > > - prot |= p->flags; > > - > > - /* and since the content will be modified, we must > invalidate > > + /* Since the content will be modified, we must > invalidate > > the corresponding translated code. */ > > if (tb_invalidate_phys_page(addr, pc)) { > > mmap_unlock(); > > @@ -2036,6 +2031,16 @@ int page_unprotect(target_ulong address, > uintptr_t pc) > > tb_invalidate_check(addr); > > #endif > > } > > + > > + /* If we got here, current TB have been retranslated > (in case of > > + * self-modifying code), now it's safe to remove page > protection. > > + */ > > + prot = 0; > > + for (addr = host_start ; addr < host_end ; addr += > TARGET_PAGE_SIZE) { > > + p = page_find(addr >> TARGET_PAGE_BITS); > > + p->flags |= PAGE_WRITE; > > + prot |= p->flags; > > + } > > mprotect((void *)g2h(host_start), qemu_host_page_size, > > prot & PAGE_BITS); > > > >
Yes, exactly. There is no point for returning to main loop immediately when current TB is found on host page and is retranslated. We can continue invalidation of TBs, and finally remove host page write protection. So there will be no second SEGFAULT. And when generating TB for next instructions, host page will be locked again, if TB includes instructions from that page. Stanislav. 6 Июл 2016 г. 15:53 пользователь "Sergey Fedorov" <sergey.fedorov@linaro.org> написал: > On 06/07/16 15:01, Стас Шмаров wrote: > > May be this can be done faster. > After invalidation of current TB, we know next executed TB will be 1 > instruction long and will write to same address, and will cause segfault. > In sigfault handler write protection to this page will be disabled. > So, we can retranslate TB, and remove protection before execution of new > TB. It will be one signal less. > > > You mean not to return 2 immediately but rather to remember the desired > return value and keep invalidating guest pages, then go ahead and call > mprotect() to enable writes to the host page? > > Thanks, > Sergey > > > 2016-07-06 13:21 GMT+03:00 Sergey Fedorov <sergey.fedorov@linaro.org>: > >> On 06/07/16 10:54, Stanislav Shmarov wrote: >> > In user-mode emulation Translation Block can consist of 2 guest pages. >> > In that case QEMU also mprotects 2 host pages that are dedicated for >> > guest memory, containing instructions. QEMU detects self-modifying code >> > with SEGFAULT signal processing. >> > >> > In case if instruction in 1st page is modifying memory of 2nd >> > page (or vice versa) QEMU will mark 2nd page with PAGE_WRITE, >> > invalidate TB, generate new TB contatining 1 guest instruction and >> > exit to CPU loop. QEMU won't call mprotect, and new TB will cause >> > same SEGFAULT. Page will have both PAGE_WRITE_ORG and PAGE_WRITE >> > flags, so QEMU will handle the signal as guest binary problem, >> > and exit with guest SEGFAULT. >> > >> > Solution is retranslate TB before marking pages as PAGE_WRITE, >> > and remove protection with mprotect on second SEGFAULT. >> > >> > Signed-off-by: Stanislav Shmarov <snarpix@gmail.com> >> >> Reviewed-by: Sergey Fedorov <sergey.fedorov@linaro.org> >> >> > --- >> > v2: Moved setting PAGE_WRITE flag to separte loop, to cover cases, >> > pointed by Sergey Fedorov. >> > >> > translate-all.c | 17 +++++++++++------ >> > 1 file changed, 11 insertions(+), 6 deletions(-) >> > >> > diff --git a/translate-all.c b/translate-all.c >> > index eaa95e4..fb3743f 100644 >> > --- a/translate-all.c >> > +++ b/translate-all.c >> > @@ -2020,13 +2020,8 @@ int page_unprotect(target_ulong address, >> uintptr_t pc) >> > host_start = address & qemu_host_page_mask; >> > host_end = host_start + qemu_host_page_size; >> > >> > - prot = 0; >> > for (addr = host_start ; addr < host_end ; addr += >> TARGET_PAGE_SIZE) { >> > - p = page_find(addr >> TARGET_PAGE_BITS); >> > - p->flags |= PAGE_WRITE; >> > - prot |= p->flags; >> > - >> > - /* and since the content will be modified, we must >> invalidate >> > + /* Since the content will be modified, we must invalidate >> > the corresponding translated code. */ >> > if (tb_invalidate_phys_page(addr, pc)) { >> > mmap_unlock(); >> > @@ -2036,6 +2031,16 @@ int page_unprotect(target_ulong address, >> uintptr_t pc) >> > tb_invalidate_check(addr); >> > #endif >> > } >> > + >> > + /* If we got here, current TB have been retranslated (in case >> of >> > + * self-modifying code), now it's safe to remove page >> protection. >> > + */ >> > + prot = 0; >> > + for (addr = host_start ; addr < host_end ; addr += >> TARGET_PAGE_SIZE) { >> > + p = page_find(addr >> TARGET_PAGE_BITS); >> > + p->flags |= PAGE_WRITE; >> > + prot |= p->flags; >> > + } >> > mprotect((void *)g2h(host_start), qemu_host_page_size, >> > prot & PAGE_BITS); >> > >> >> > >
On 06/07/16 16:22, Stanislav Shmarov wrote: > > Yes, exactly. > > There is no point for returning to main loop immediately when current > TB is found on host page and is retranslated. We can continue > invalidation of TBs, and finally remove host page write protection. So > there will be no second SEGFAULT. > > And when generating TB for next instructions, host page will be locked > again, if TB includes instructions from that page. > I think that could work. Regards, Sergey
diff --git a/translate-all.c b/translate-all.c index eaa95e4..fb3743f 100644 --- a/translate-all.c +++ b/translate-all.c @@ -2020,13 +2020,8 @@ int page_unprotect(target_ulong address, uintptr_t pc) host_start = address & qemu_host_page_mask; host_end = host_start + qemu_host_page_size; - prot = 0; for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) { - p = page_find(addr >> TARGET_PAGE_BITS); - p->flags |= PAGE_WRITE; - prot |= p->flags; - - /* and since the content will be modified, we must invalidate + /* Since the content will be modified, we must invalidate the corresponding translated code. */ if (tb_invalidate_phys_page(addr, pc)) { mmap_unlock(); @@ -2036,6 +2031,16 @@ int page_unprotect(target_ulong address, uintptr_t pc) tb_invalidate_check(addr); #endif } + + /* If we got here, current TB have been retranslated (in case of + * self-modifying code), now it's safe to remove page protection. + */ + prot = 0; + for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) { + p = page_find(addr >> TARGET_PAGE_BITS); + p->flags |= PAGE_WRITE; + prot |= p->flags; + } mprotect((void *)g2h(host_start), qemu_host_page_size, prot & PAGE_BITS);
In user-mode emulation Translation Block can consist of 2 guest pages. In that case QEMU also mprotects 2 host pages that are dedicated for guest memory, containing instructions. QEMU detects self-modifying code with SEGFAULT signal processing. In case if instruction in 1st page is modifying memory of 2nd page (or vice versa) QEMU will mark 2nd page with PAGE_WRITE, invalidate TB, generate new TB contatining 1 guest instruction and exit to CPU loop. QEMU won't call mprotect, and new TB will cause same SEGFAULT. Page will have both PAGE_WRITE_ORG and PAGE_WRITE flags, so QEMU will handle the signal as guest binary problem, and exit with guest SEGFAULT. Solution is retranslate TB before marking pages as PAGE_WRITE, and remove protection with mprotect on second SEGFAULT. Signed-off-by: Stanislav Shmarov <snarpix@gmail.com> --- v2: Moved setting PAGE_WRITE flag to separte loop, to cover cases, pointed by Sergey Fedorov. translate-all.c | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-)