| Message ID | 20211213112034.2896536-1-heiko@sntech.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] riscv: prevent null-pointer dereference with sbi_remote_fence_i | expand |
On Mon, Dec 13, 2021 at 3:21 AM Heiko Stuebner <heiko@sntech.de> wrote: > > The callback used inside sbi_remote_fence_i is set at sbi probe time > to the needed variant. Before that it is a NULL pointer. > > The selection between using sbi_remote_fence_i or ipi_remote_fence_i > right now is solely done on the presence of the RISCV_SBI config option. > > On a multiplatform kernel, SBI will probably always be built in, but > in the future not all machines using that kernel may have SBI > on them, so in some cases this setup can lead to NULL pointer dereferences. > I didn't quite understand this. The only platforms without SBI are M-mode Linux based platforms. All other platforms will require SBI if the kernel is running on supervisor mode. They may not need SBI IPI or SBI RFENCE extensions if an alternate mechanism(ACLINT SSWI or AIA IMSIC) is already available in hardware. IIRC, Anup's aclint & aia series already takes care of that. > Also if in future one of the flush_icache_all / flush_icache_mm functions > gets called earlier in the boot process - before sbi_init - this would > trigger the issue. > sbi_init is called before setup_smp. Do you think there is a use case where flush_icache_xxx can be called before smp initialization ? > To prevent this, add a default __sbi_rfence_none returning an error code > and adapt the callers to check for an error from the remote fence > to then fall back to the other method. > Having said that, it is always good to have guards to avoid NULL pointer dereferences. However, the commit text may be updated based on the above questions. > Signed-off-by: Heiko Stuebner <heiko@sntech.de> > --- > arch/riscv/kernel/sbi.c | 10 +++++++++- > arch/riscv/mm/cacheflush.c | 25 +++++++++++++++++-------- > 2 files changed, 26 insertions(+), 9 deletions(-) > > diff --git a/arch/riscv/kernel/sbi.c b/arch/riscv/kernel/sbi.c > index 7402a417f38e..69d0a96b97d0 100644 > --- a/arch/riscv/kernel/sbi.c > +++ b/arch/riscv/kernel/sbi.c > @@ -14,11 +14,19 @@ > unsigned long sbi_spec_version __ro_after_init = SBI_SPEC_VERSION_DEFAULT; > EXPORT_SYMBOL(sbi_spec_version); > > +static int __sbi_rfence_none(int fid, const unsigned long *hart_mask, > + unsigned long start, unsigned long size, > + unsigned long arg4, unsigned long arg5) > +{ > + return -EOPNOTSUPP; > +} > + > static void (*__sbi_set_timer)(uint64_t stime) __ro_after_init; > static int (*__sbi_send_ipi)(const unsigned long *hart_mask) __ro_after_init; > static int (*__sbi_rfence)(int fid, const unsigned long *hart_mask, > unsigned long start, unsigned long size, > - unsigned long arg4, unsigned long arg5) __ro_after_init; > + unsigned long arg4, unsigned long arg5) > + __ro_after_init = __sbi_rfence_none; > > struct sbiret sbi_ecall(int ext, int fid, unsigned long arg0, > unsigned long arg1, unsigned long arg2, > diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c > index 89f81067e09e..128e23c094ea 100644 > --- a/arch/riscv/mm/cacheflush.c > +++ b/arch/riscv/mm/cacheflush.c > @@ -16,11 +16,15 @@ static void ipi_remote_fence_i(void *info) > > void flush_icache_all(void) > { > + int ret = -EINVAL; > + > local_flush_icache_all(); > > if (IS_ENABLED(CONFIG_RISCV_SBI)) > - sbi_remote_fence_i(NULL); > - else > + ret = sbi_remote_fence_i(NULL); > + > + /* fall back to ipi_remote_fence_i if sbi failed or not available */ > + if (ret) > on_each_cpu(ipi_remote_fence_i, NULL, 1); > } > EXPORT_SYMBOL(flush_icache_all); > @@ -66,13 +70,18 @@ void flush_icache_mm(struct mm_struct *mm, bool local) > * with flush_icache_deferred(). > */ > smp_mb(); > - } else if (IS_ENABLED(CONFIG_RISCV_SBI)) { > - cpumask_t hartid_mask; > - > - riscv_cpuid_to_hartid_mask(&others, &hartid_mask); > - sbi_remote_fence_i(cpumask_bits(&hartid_mask)); > } else { > - on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); > + int ret = -EINVAL; > + > + if (IS_ENABLED(CONFIG_RISCV_SBI)) { > + cpumask_t hartid_mask; > + > + riscv_cpuid_to_hartid_mask(&others, &hartid_mask); > + ret = sbi_remote_fence_i(cpumask_bits(&hartid_mask)); This API is removed in sparse hartid series[1]. You may want to rebase on top of it. [1] https://patchwork.kernel.org/project/linux-riscv/list/?series=590195 > + } > + > + if (ret) > + on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); > } > > preempt_enable(); > -- > 2.30.2 > > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv
Hi Atish, Am Dienstag, 14. Dezember 2021, 02:51:17 CET schrieb Atish Patra: > On Mon, Dec 13, 2021 at 3:21 AM Heiko Stuebner <heiko@sntech.de> wrote: > > > > The callback used inside sbi_remote_fence_i is set at sbi probe time > > to the needed variant. Before that it is a NULL pointer. > > > > The selection between using sbi_remote_fence_i or ipi_remote_fence_i > > right now is solely done on the presence of the RISCV_SBI config option. > > > > On a multiplatform kernel, SBI will probably always be built in, but > > in the future not all machines using that kernel may have SBI > > on them, so in some cases this setup can lead to NULL pointer dereferences. > > > > I didn't quite understand this. The only platforms without SBI are > M-mode Linux based platforms. > All other platforms will require SBI if the kernel is running on > supervisor mode. Ok ... but imagine a single-core board starting in S-mode. Not doing smp, kvm, etc. So I can very well imagine the possibility that people might start it without an sbi instance running ;-) . > They may not need SBI IPI or SBI RFENCE extensions if an alternate > mechanism(ACLINT SSWI or AIA IMSIC) > is already available in hardware. IIRC, Anup's aclint & aia series > already takes care of that. > > > Also if in future one of the flush_icache_all / flush_icache_mm functions > > gets called earlier in the boot process - before sbi_init - this would > > trigger the issue. > > > > sbi_init is called before setup_smp. Do you think there is a use case where > flush_icache_xxx can be called before smp initialization ? I ran into the issue while experimenting with doing alternatives patching earlier. Works fine, but the flush_icache call then runs into that null-ptr. Heiko > > To prevent this, add a default __sbi_rfence_none returning an error code > > and adapt the callers to check for an error from the remote fence > > to then fall back to the other method. > > > > Having said that, it is always good to have guards to avoid NULL > pointer dereferences. > However, the commit text may be updated based on the above questions. > > > Signed-off-by: Heiko Stuebner <heiko@sntech.de> > > --- > > arch/riscv/kernel/sbi.c | 10 +++++++++- > > arch/riscv/mm/cacheflush.c | 25 +++++++++++++++++-------- > > 2 files changed, 26 insertions(+), 9 deletions(-) > > > > diff --git a/arch/riscv/kernel/sbi.c b/arch/riscv/kernel/sbi.c > > index 7402a417f38e..69d0a96b97d0 100644 > > --- a/arch/riscv/kernel/sbi.c > > +++ b/arch/riscv/kernel/sbi.c > > @@ -14,11 +14,19 @@ > > unsigned long sbi_spec_version __ro_after_init = SBI_SPEC_VERSION_DEFAULT; > > EXPORT_SYMBOL(sbi_spec_version); > > > > +static int __sbi_rfence_none(int fid, const unsigned long *hart_mask, > > + unsigned long start, unsigned long size, > > + unsigned long arg4, unsigned long arg5) > > +{ > > + return -EOPNOTSUPP; > > +} > > + > > static void (*__sbi_set_timer)(uint64_t stime) __ro_after_init; > > static int (*__sbi_send_ipi)(const unsigned long *hart_mask) __ro_after_init; > > static int (*__sbi_rfence)(int fid, const unsigned long *hart_mask, > > unsigned long start, unsigned long size, > > - unsigned long arg4, unsigned long arg5) __ro_after_init; > > + unsigned long arg4, unsigned long arg5) > > + __ro_after_init = __sbi_rfence_none; > > > > struct sbiret sbi_ecall(int ext, int fid, unsigned long arg0, > > unsigned long arg1, unsigned long arg2, > > diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c > > index 89f81067e09e..128e23c094ea 100644 > > --- a/arch/riscv/mm/cacheflush.c > > +++ b/arch/riscv/mm/cacheflush.c > > @@ -16,11 +16,15 @@ static void ipi_remote_fence_i(void *info) > > > > void flush_icache_all(void) > > { > > + int ret = -EINVAL; > > + > > local_flush_icache_all(); > > > > if (IS_ENABLED(CONFIG_RISCV_SBI)) > > - sbi_remote_fence_i(NULL); > > - else > > + ret = sbi_remote_fence_i(NULL); > > + > > + /* fall back to ipi_remote_fence_i if sbi failed or not available */ > > + if (ret) > > on_each_cpu(ipi_remote_fence_i, NULL, 1); > > } > > EXPORT_SYMBOL(flush_icache_all); > > @@ -66,13 +70,18 @@ void flush_icache_mm(struct mm_struct *mm, bool local) > > * with flush_icache_deferred(). > > */ > > smp_mb(); > > - } else if (IS_ENABLED(CONFIG_RISCV_SBI)) { > > - cpumask_t hartid_mask; > > - > > - riscv_cpuid_to_hartid_mask(&others, &hartid_mask); > > - sbi_remote_fence_i(cpumask_bits(&hartid_mask)); > > } else { > > - on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); > > + int ret = -EINVAL; > > + > > + if (IS_ENABLED(CONFIG_RISCV_SBI)) { > > + cpumask_t hartid_mask; > > + > > + riscv_cpuid_to_hartid_mask(&others, &hartid_mask); > > + ret = sbi_remote_fence_i(cpumask_bits(&hartid_mask)); > > This API is removed in sparse hartid series[1]. You may want to rebase > on top of it. > > [1] https://patchwork.kernel.org/project/linux-riscv/list/?series=590195 > > > + } > > + > > + if (ret) > > + on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); > > } > > > > preempt_enable(); > > -- > > 2.30.2 > > > > > > _______________________________________________ > > linux-riscv mailing list > > linux-riscv@lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-riscv > > > >
diff --git a/arch/riscv/kernel/sbi.c b/arch/riscv/kernel/sbi.c index 7402a417f38e..69d0a96b97d0 100644 --- a/arch/riscv/kernel/sbi.c +++ b/arch/riscv/kernel/sbi.c @@ -14,11 +14,19 @@ unsigned long sbi_spec_version __ro_after_init = SBI_SPEC_VERSION_DEFAULT; EXPORT_SYMBOL(sbi_spec_version); +static int __sbi_rfence_none(int fid, const unsigned long *hart_mask, + unsigned long start, unsigned long size, + unsigned long arg4, unsigned long arg5) +{ + return -EOPNOTSUPP; +} + static void (*__sbi_set_timer)(uint64_t stime) __ro_after_init; static int (*__sbi_send_ipi)(const unsigned long *hart_mask) __ro_after_init; static int (*__sbi_rfence)(int fid, const unsigned long *hart_mask, unsigned long start, unsigned long size, - unsigned long arg4, unsigned long arg5) __ro_after_init; + unsigned long arg4, unsigned long arg5) + __ro_after_init = __sbi_rfence_none; struct sbiret sbi_ecall(int ext, int fid, unsigned long arg0, unsigned long arg1, unsigned long arg2, diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c index 89f81067e09e..128e23c094ea 100644 --- a/arch/riscv/mm/cacheflush.c +++ b/arch/riscv/mm/cacheflush.c @@ -16,11 +16,15 @@ static void ipi_remote_fence_i(void *info) void flush_icache_all(void) { + int ret = -EINVAL; + local_flush_icache_all(); if (IS_ENABLED(CONFIG_RISCV_SBI)) - sbi_remote_fence_i(NULL); - else + ret = sbi_remote_fence_i(NULL); + + /* fall back to ipi_remote_fence_i if sbi failed or not available */ + if (ret) on_each_cpu(ipi_remote_fence_i, NULL, 1); } EXPORT_SYMBOL(flush_icache_all); @@ -66,13 +70,18 @@ void flush_icache_mm(struct mm_struct *mm, bool local) * with flush_icache_deferred(). */ smp_mb(); - } else if (IS_ENABLED(CONFIG_RISCV_SBI)) { - cpumask_t hartid_mask; - - riscv_cpuid_to_hartid_mask(&others, &hartid_mask); - sbi_remote_fence_i(cpumask_bits(&hartid_mask)); } else { - on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); + int ret = -EINVAL; + + if (IS_ENABLED(CONFIG_RISCV_SBI)) { + cpumask_t hartid_mask; + + riscv_cpuid_to_hartid_mask(&others, &hartid_mask); + ret = sbi_remote_fence_i(cpumask_bits(&hartid_mask)); + } + + if (ret) + on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1); } preempt_enable();
The callback used inside sbi_remote_fence_i is set at sbi probe time to the needed variant. Before that it is a NULL pointer. The selection between using sbi_remote_fence_i or ipi_remote_fence_i right now is solely done on the presence of the RISCV_SBI config option. On a multiplatform kernel, SBI will probably always be built in, but in the future not all machines using that kernel may have SBI on them, so in some cases this setup can lead to NULL pointer dereferences. Also if in future one of the flush_icache_all / flush_icache_mm functions gets called earlier in the boot process - before sbi_init - this would trigger the issue. To prevent this, add a default __sbi_rfence_none returning an error code and adapt the callers to check for an error from the remote fence to then fall back to the other method. Signed-off-by: Heiko Stuebner <heiko@sntech.de> --- arch/riscv/kernel/sbi.c | 10 +++++++++- arch/riscv/mm/cacheflush.c | 25 +++++++++++++++++-------- 2 files changed, 26 insertions(+), 9 deletions(-)