| Message ID | 20220414173037.1381927-1-niklas.cassel@wdc.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | riscv: Don't output a bogus mmu-type on a no MMU kernel | expand |
On Thu, 14 Apr 2022 10:30:36 PDT (-0700), niklas.cassel@wdc.com wrote: > Currently on a 64-bit kernel built without CONFIG_MMU, /proc/cpuinfo will > show the current MMU mode as sv57. > > While the device tree property "mmu-type" does have a value "riscv,none" to > describe a CPU without a MMU, since commit 73c7c8f68e72 ("riscv: Use > pgtable_l4_enabled to output mmu_type in cpuinfo"), we no longer rely on > device tree to output the MMU mode. (Not even for CONFIG_32BIT.) > > Therefore, instead of readding code to look at the "mmu-type" device tree > property, let's continue with the existing convention to use fixed values > for configurations where we don't determine the MMU mode at runtime. > > Add a new fixed value for !CONFIG_MMU in order to output the correct > MMU mode in cpuinfo. There's really two ideas as to what /proc/cpuinfo should be: do we show what the HW has, or what we userspace sees. This sort of thing is a perfect example of that split. We've been kind of vague about this in the past, but IMO putting what userspace sees in /proc/cpuinfo (and HWCAP, etc) is the right way to go. That does hide a bit from userspace WRT what hardware it's running on, but it's more in line with the design of RISC-V (ie, a lot is hidden from userspace). I've put this on for-next. Thanks! > Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com> > --- > arch/riscv/kernel/cpu.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/arch/riscv/kernel/cpu.c b/arch/riscv/kernel/cpu.c > index ccb617791e56..ecfb3e85ffb2 100644 > --- a/arch/riscv/kernel/cpu.c > +++ b/arch/riscv/kernel/cpu.c > @@ -138,6 +138,7 @@ static void print_mmu(struct seq_file *f) > { > char sv_type[16]; > > +#ifdef CONFIG_MMU > #if defined(CONFIG_32BIT) > strncpy(sv_type, "sv32", 5); > #elif defined(CONFIG_64BIT) > @@ -148,6 +149,9 @@ static void print_mmu(struct seq_file *f) > else > strncpy(sv_type, "sv39", 5); > #endif > +#else > + strncpy(sv_type, "none", 5); > +#endif /* CONFIG_MMU */ > seq_printf(f, "mmu\t\t: %s\n", sv_type); > }
On Sat, May 21, 2022 at 1:47 PM Palmer Dabbelt <palmer@dabbelt.com> wrote: > > On Thu, 14 Apr 2022 10:30:36 PDT (-0700), niklas.cassel@wdc.com wrote: > > Currently on a 64-bit kernel built without CONFIG_MMU, /proc/cpuinfo will > > show the current MMU mode as sv57. > > > > While the device tree property "mmu-type" does have a value "riscv,none" to > > describe a CPU without a MMU, since commit 73c7c8f68e72 ("riscv: Use > > pgtable_l4_enabled to output mmu_type in cpuinfo"), we no longer rely on > > device tree to output the MMU mode. (Not even for CONFIG_32BIT.) > > > > Therefore, instead of readding code to look at the "mmu-type" device tree > > property, let's continue with the existing convention to use fixed values > > for configurations where we don't determine the MMU mode at runtime. > > > > Add a new fixed value for !CONFIG_MMU in order to output the correct > > MMU mode in cpuinfo. > > There's really two ideas as to what /proc/cpuinfo should be: do we show > what the HW has, or what we userspace sees. This sort of thing is a > perfect example of that split. We've been kind of vague about this in > the past, but IMO putting what userspace sees in /proc/cpuinfo (and > HWCAP, etc) is the right way to go. That does hide a bit from userspace > WRT what hardware it's running on, but it's more in line with the design > of RISC-V (ie, a lot is hidden from userspace). > How about vendor id, arch id & implementation id ? Looking at the /proc/cpuinfo output on arm64/x86, it seems to print related information there. --------------------------------------------------- /proc/cpuinfo output in x86 --------------------------------------------------- processor : 31 vendor_id : GenuineIntel cpu family : 6 model : 85 model name : Intel(R) Xeon(R) Gold 6246R CPU @ 3.40GHz --------------------------------------------------- /proc/cpuinfo output in aarm64 --------------------------------------------------- ubuntu@ubuntu:~$ cat /proc/cpuinfo processor : 0 CPU implementer : 0x41 CPU architecture: 8 CPU variant : 0x1 CPU part : 0xd07 CPU revision : 0 --------------------------------------------------- > I've put this on for-next. > > Thanks! > > > Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com> > > --- > > arch/riscv/kernel/cpu.c | 4 ++++ > > 1 file changed, 4 insertions(+) > > > > diff --git a/arch/riscv/kernel/cpu.c b/arch/riscv/kernel/cpu.c > > index ccb617791e56..ecfb3e85ffb2 100644 > > --- a/arch/riscv/kernel/cpu.c > > +++ b/arch/riscv/kernel/cpu.c > > @@ -138,6 +138,7 @@ static void print_mmu(struct seq_file *f) > > { > > char sv_type[16]; > > > > +#ifdef CONFIG_MMU > > #if defined(CONFIG_32BIT) > > strncpy(sv_type, "sv32", 5); > > #elif defined(CONFIG_64BIT) > > @@ -148,6 +149,9 @@ static void print_mmu(struct seq_file *f) > > else > > strncpy(sv_type, "sv39", 5); > > #endif > > +#else > > + strncpy(sv_type, "none", 5); > > +#endif /* CONFIG_MMU */ > > seq_printf(f, "mmu\t\t: %s\n", sv_type); > > } > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv
On Sat, May 21, 2022 at 01:46:43PM -0700, Palmer Dabbelt wrote: > On Thu, 14 Apr 2022 10:30:36 PDT (-0700), niklas.cassel@wdc.com wrote: > > Currently on a 64-bit kernel built without CONFIG_MMU, /proc/cpuinfo will > > show the current MMU mode as sv57. > > > > While the device tree property "mmu-type" does have a value "riscv,none" to > > describe a CPU without a MMU, since commit 73c7c8f68e72 ("riscv: Use > > pgtable_l4_enabled to output mmu_type in cpuinfo"), we no longer rely on > > device tree to output the MMU mode. (Not even for CONFIG_32BIT.) > > > > Therefore, instead of readding code to look at the "mmu-type" device tree > > property, let's continue with the existing convention to use fixed values > > for configurations where we don't determine the MMU mode at runtime. > > > > Add a new fixed value for !CONFIG_MMU in order to output the correct > > MMU mode in cpuinfo. > > There's really two ideas as to what /proc/cpuinfo should be: do we show what > the HW has, or what we userspace sees. This sort of thing is a perfect > example of that split. We've been kind of vague about this in the past, but > IMO putting what userspace sees in /proc/cpuinfo (and HWCAP, etc) is the > right way to go. That does hide a bit from userspace WRT what hardware it's > running on, but it's more in line with the design of RISC-V (ie, a lot is > hidden from userspace). > > I've put this on for-next. Thank you Palmer! I did send out (somewhat) related device tree binding fix: https://lore.kernel.org/linux-riscv/YoH9TE%2F4ruFQw3fV@x1-carbon/T/#m894179b986b2e1ef8ef43dd4a865f872e2564c18 Any chance of that getting picked up as well? Kind regards, Niklas
diff --git a/arch/riscv/kernel/cpu.c b/arch/riscv/kernel/cpu.c index ccb617791e56..ecfb3e85ffb2 100644 --- a/arch/riscv/kernel/cpu.c +++ b/arch/riscv/kernel/cpu.c @@ -138,6 +138,7 @@ static void print_mmu(struct seq_file *f) { char sv_type[16]; +#ifdef CONFIG_MMU #if defined(CONFIG_32BIT) strncpy(sv_type, "sv32", 5); #elif defined(CONFIG_64BIT) @@ -148,6 +149,9 @@ static void print_mmu(struct seq_file *f) else strncpy(sv_type, "sv39", 5); #endif +#else + strncpy(sv_type, "none", 5); +#endif /* CONFIG_MMU */ seq_printf(f, "mmu\t\t: %s\n", sv_type); }
Currently on a 64-bit kernel built without CONFIG_MMU, /proc/cpuinfo will show the current MMU mode as sv57. While the device tree property "mmu-type" does have a value "riscv,none" to describe a CPU without a MMU, since commit 73c7c8f68e72 ("riscv: Use pgtable_l4_enabled to output mmu_type in cpuinfo"), we no longer rely on device tree to output the MMU mode. (Not even for CONFIG_32BIT.) Therefore, instead of readding code to look at the "mmu-type" device tree property, let's continue with the existing convention to use fixed values for configurations where we don't determine the MMU mode at runtime. Add a new fixed value for !CONFIG_MMU in order to output the correct MMU mode in cpuinfo. Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com> --- arch/riscv/kernel/cpu.c | 4 ++++ 1 file changed, 4 insertions(+)