Message ID | 20241025175857.2554252-2-pierrick.bouvier@linaro.org |
---|---|
State | Superseded |
Headers | show |
Series | target/i386: fix hang when using slow path for ptw_setl | expand |
On 10/25/24 18:58, Pierrick Bouvier wrote: > When instrumenting memory accesses for plugin, we force memory accesses > to use the slow path for mmu [1]. This create a situation where we end > up calling ptw_setl_slow. This was fixed recently in [2] but the issue > still could appear out of plugins use case. > > Since this function gets called during a cpu_exec, start_exclusive then > hangs. This exclusive section was introduced initially for security > reasons [3]. > > I suspect this code path was never triggered, because ptw_setl_slow > would always be called transitively from cpu_exec, resulting in a hang. > > [1]https://gitlab.com/qemu-project/qemu/-/commit/6d03226b42247b68ab2f0b3663e0f624335a4055 > [2]https://gitlab.com/qemu-project/qemu/-/commit/115ade42d50144c15b74368d32dc734ea277d853 > [3]https://gitlab.com/qemu-project/qemu/-/issues/279 > > Fixes:https://gitlab.com/qemu-project/qemu/-/issues/2566 > Signed-off-by: Pierrick Bouvier<pierrick.bouvier@linaro.org> > --- > target/i386/tcg/sysemu/excp_helper.c | 5 +++++ > 1 file changed, 5 insertions(+) Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
25.10.2024 20:58, Pierrick Bouvier wrote: > When instrumenting memory accesses for plugin, we force memory accesses > to use the slow path for mmu [1]. This create a situation where we end > up calling ptw_setl_slow. This was fixed recently in [2] but the issue > still could appear out of plugins use case. > > Since this function gets called during a cpu_exec, start_exclusive then > hangs. This exclusive section was introduced initially for security > reasons [3]. > > I suspect this code path was never triggered, because ptw_setl_slow > would always be called transitively from cpu_exec, resulting in a hang. > > [1] https://gitlab.com/qemu-project/qemu/-/commit/6d03226b42247b68ab2f0b3663e0f624335a4055 > [2] https://gitlab.com/qemu-project/qemu/-/commit/115ade42d50144c15b74368d32dc734ea277d853 > [3] https://gitlab.com/qemu-project/qemu/-/issues/279 > > Fixes: https://gitlab.com/qemu-project/qemu/-/issues/2566 > Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> [1] is in 8.2.x. [2] is in 9.2.tobe, and marked as should be picked up for stable (I picked it up for 8.2.x, 9.0.x and 9.1.x). Shouldn't this one be picked up for stable too, as an addition fix ontop of [2]? Or is it not important? (I guess since it's reported in our issue tracker, it is problematic for someone already). I picked this one up for 8.2, 9.0 and 9.1 stable series -- please let me know if I should not. Also, what about the 2/2 in this series, "cpu: ensure we don't call start_exclusive from cpu_exec", which is 779f30a01af8566780cefc8639505b758950afb3 in master now? Thanks, /mjt > target/i386/tcg/sysemu/excp_helper.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/target/i386/tcg/sysemu/excp_helper.c b/target/i386/tcg/sysemu/excp_helper.c > index da187c8792a..ddc51e3e0b8 100644 > --- a/target/i386/tcg/sysemu/excp_helper.c > +++ b/target/i386/tcg/sysemu/excp_helper.c > @@ -107,6 +107,10 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) > { > uint32_t cmp; > > + CPUState *cpu = env_cpu(in->env); > + /* We are in cpu_exec, and start_exclusive can't be called directly.*/ > + g_assert(cpu->running); > + cpu_exec_end(cpu); > /* Does x86 really perform a rmw cycle on mmio for ptw? */ > start_exclusive(); > cmp = cpu_ldl_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); > @@ -114,6 +118,7 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) > cpu_stl_mmuidx_ra(in->env, in->gaddr, new, in->ptw_idx, 0); > } > end_exclusive(); > + cpu_exec_start(cpu); > return cmp == old; > } >
Hi Michael, On 11/16/24 19:49, Michael Tokarev wrote: > 25.10.2024 20:58, Pierrick Bouvier wrote: >> When instrumenting memory accesses for plugin, we force memory accesses >> to use the slow path for mmu [1]. This create a situation where we end >> up calling ptw_setl_slow. This was fixed recently in [2] but the issue >> still could appear out of plugins use case. >> >> Since this function gets called during a cpu_exec, start_exclusive then >> hangs. This exclusive section was introduced initially for security >> reasons [3]. >> >> I suspect this code path was never triggered, because ptw_setl_slow >> would always be called transitively from cpu_exec, resulting in a hang. >> >> [1] https://gitlab.com/qemu-project/qemu/-/commit/6d03226b42247b68ab2f0b3663e0f624335a4055 >> [2] https://gitlab.com/qemu-project/qemu/-/commit/115ade42d50144c15b74368d32dc734ea277d853 >> [3] https://gitlab.com/qemu-project/qemu/-/issues/279 >> >> Fixes: https://gitlab.com/qemu-project/qemu/-/issues/2566 >> Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> > > [1] is in 8.2.x. [2] is in 9.2.tobe, and marked as should be picked up > for stable (I picked it up for 8.2.x, 9.0.x and 9.1.x). > > Shouldn't this one be picked up for stable too, as an addition fix > ontop of [2]? Or is it not important? (I guess since it's reported > in our issue tracker, it is problematic for someone already). > > I picked this one up for 8.2, 9.0 and 9.1 stable series -- please > let me know if I should not. > You can safely integrate this on all stable branches. It will fix the issue with plugins, and might possibly fix an issue with the memory access if someone hit that path. The original problematic change was missing this fix from the start. > Also, what about the 2/2 in this series, "cpu: ensure we don't call > start_exclusive from cpu_exec", which is 779f30a01af8566780cefc8639505b758950afb3 > in master now? > It's an assert to ensure we don't hit the same blocking situation in the future (with new changes), so I don't think there is an added value to integrate that on stable branches. Thanks, Pierrick > Thanks, > > /mjt > >> target/i386/tcg/sysemu/excp_helper.c | 5 +++++ >> 1 file changed, 5 insertions(+) >> >> diff --git a/target/i386/tcg/sysemu/excp_helper.c b/target/i386/tcg/sysemu/excp_helper.c >> index da187c8792a..ddc51e3e0b8 100644 >> --- a/target/i386/tcg/sysemu/excp_helper.c >> +++ b/target/i386/tcg/sysemu/excp_helper.c >> @@ -107,6 +107,10 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) >> { >> uint32_t cmp; >> >> + CPUState *cpu = env_cpu(in->env); >> + /* We are in cpu_exec, and start_exclusive can't be called directly.*/ >> + g_assert(cpu->running); >> + cpu_exec_end(cpu); >> /* Does x86 really perform a rmw cycle on mmio for ptw? */ >> start_exclusive(); >> cmp = cpu_ldl_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); >> @@ -114,6 +118,7 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) >> cpu_stl_mmuidx_ra(in->env, in->gaddr, new, in->ptw_idx, 0); >> } >> end_exclusive(); >> + cpu_exec_start(cpu); >> return cmp == old; >> } >> > >
diff --git a/target/i386/tcg/sysemu/excp_helper.c b/target/i386/tcg/sysemu/excp_helper.c index da187c8792a..ddc51e3e0b8 100644 --- a/target/i386/tcg/sysemu/excp_helper.c +++ b/target/i386/tcg/sysemu/excp_helper.c @@ -107,6 +107,10 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) { uint32_t cmp; + CPUState *cpu = env_cpu(in->env); + /* We are in cpu_exec, and start_exclusive can't be called directly.*/ + g_assert(cpu->running); + cpu_exec_end(cpu); /* Does x86 really perform a rmw cycle on mmio for ptw? */ start_exclusive(); cmp = cpu_ldl_mmuidx_ra(in->env, in->gaddr, in->ptw_idx, 0); @@ -114,6 +118,7 @@ static bool ptw_setl_slow(const PTETranslate *in, uint32_t old, uint32_t new) cpu_stl_mmuidx_ra(in->env, in->gaddr, new, in->ptw_idx, 0); } end_exclusive(); + cpu_exec_start(cpu); return cmp == old; }
When instrumenting memory accesses for plugin, we force memory accesses to use the slow path for mmu [1]. This create a situation where we end up calling ptw_setl_slow. This was fixed recently in [2] but the issue still could appear out of plugins use case. Since this function gets called during a cpu_exec, start_exclusive then hangs. This exclusive section was introduced initially for security reasons [3]. I suspect this code path was never triggered, because ptw_setl_slow would always be called transitively from cpu_exec, resulting in a hang. [1] https://gitlab.com/qemu-project/qemu/-/commit/6d03226b42247b68ab2f0b3663e0f624335a4055 [2] https://gitlab.com/qemu-project/qemu/-/commit/115ade42d50144c15b74368d32dc734ea277d853 [3] https://gitlab.com/qemu-project/qemu/-/issues/279 Fixes: https://gitlab.com/qemu-project/qemu/-/issues/2566 Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> --- target/i386/tcg/sysemu/excp_helper.c | 5 +++++ 1 file changed, 5 insertions(+)