| Message ID | 20211203141416.31121-1-mark-pk.tsai@mediatek.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm: remove [_text, _stext) from kernel code resource | expand |
On Fri, Dec 03, 2021 at 10:14:16PM +0800, Mark-PK Tsai wrote: > Remove the [_text, _stext) from Kernel Code. > Although there are some startup code in head.text, they > are freed to the buddy system after kernel boot. Hmm, I don't see it is being freed anywhere. Can you elaborate when and how the range [_text, _stext) is freed? > And we have memory protection mechanism use this > which have false alarm when some other IPs doing dma > if the dma page frame is in the [_text, _stext). > > Below are my iomem resource and reserved memory information: > console:/ # grep Kernel /proc/iomem > 20208000-219fffff : Kernel code > 21b00000-21c2e76f : Kernel data > > console:/ # cat /sys/kernel/debug/memblock/reserved > 0: 0x20201000..0x20207fff > 1: 0x20300000..0x21c2e76f What are the addresses of _text and _stext in your configuration? What these dumps are supposed to show here? > Signed-off-by: Mark-PK Tsai <mark-pk.tsai@mediatek.com> > --- > arch/arm/kernel/setup.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c > index 284a80c0b6e1..85ffc9501e38 100644 > --- a/arch/arm/kernel/setup.c > +++ b/arch/arm/kernel/setup.c > @@ -851,7 +851,7 @@ static void __init request_standard_resources(const struct machine_desc *mdesc) > struct resource *res; > u64 i; > > - kernel_code.start = virt_to_phys(_text); > + kernel_code.start = virt_to_phys(_stext); > kernel_code.end = virt_to_phys(__init_begin - 1); > kernel_data.start = virt_to_phys(_sdata); > kernel_data.end = virt_to_phys(_end - 1); > -- > 2.18.0 >
> On Fri, Dec 03, 2021 at 10:14:16PM +0800, Mark-PK Tsai wrote: > > Remove the [_text, _stext) from Kernel Code. > > Although there are some startup code in head.text, they > > are freed to the buddy system after kernel boot. > > Hmm, I don't see it is being freed anywhere. Can you elaborate when and how > the range [_text, _stext) is freed? arm_memblock_init() reserve [KERNEL_START, KERNEL_END) which are defined as following. #define KERNEL_START _stext #define KERNEL_END _end free_low_memory_core_early() free all the non-reserved range in lowmem, so the range [_text, _stext) is also freed here. > > > And we have memory protection mechanism use this > > which have false alarm when some other IPs doing dma > > if the dma page frame is in the [_text, _stext). > > > > Below are my iomem resource and reserved memory information: > > console:/ # grep Kernel /proc/iomem > > 20208000-219fffff : Kernel code > > 21b00000-21c2e76f : Kernel data > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > 0: 0x20201000..0x20207fff > > 1: 0x20300000..0x21c2e76f > > What are the addresses of _text and _stext in your configuration? va pa c0008000 20208000 _text c0100000 20300000 _stext > > What these dumps are supposed to show here? > Below is the dump info after applied this patch. console:/ # grep Kernel /proc/iomem 20300000-219fffff : Kernel code 21b00000-21c2e76f : Kernel data console:/ # cat /sys/kernel/debug/memblock/reserved 0: 0x20201000..0x20207fff 1: 0x20300000..0x21c2e76f The difference is that Kernel Code resource match the reserved memblock 1 which is reserved in arm_memblock_init(). > > > Signed-off-by: Mark-PK Tsai <mark-pk.tsai@mediatek.com> > > --- > > arch/arm/kernel/setup.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c > > index 284a80c0b6e1..85ffc9501e38 100644 > > --- a/arch/arm/kernel/setup.c > > +++ b/arch/arm/kernel/setup.c > > @@ -851,7 +851,7 @@ static void __init request_standard_resources(const struct machine_desc *mdesc) > > struct resource *res; > > u64 i; > > > > - kernel_code.start = virt_to_phys(_text); > > + kernel_code.start = virt_to_phys(_stext); > > kernel_code.end = virt_to_phys(__init_begin - 1); > > kernel_data.start = virt_to_phys(_sdata); > > kernel_data.end = virt_to_phys(_end - 1); > > -- > > 2.18.0 > > > > -- > Sincerely yours, > Mike.
On Wed, Dec 08, 2021 at 10:26:23AM +0800, Mark-PK Tsai wrote: > > On Fri, Dec 03, 2021 at 10:14:16PM +0800, Mark-PK Tsai wrote: > > > Remove the [_text, _stext) from Kernel Code. > > > Although there are some startup code in head.text, they > > > are freed to the buddy system after kernel boot. > > > > Hmm, I don't see it is being freed anywhere. Can you elaborate when and how > > the range [_text, _stext) is freed? > > arm_memblock_init() reserve [KERNEL_START, KERNEL_END) which are defined as following. > > #define KERNEL_START _stext > #define KERNEL_END _end > > free_low_memory_core_early() free all the non-reserved range in lowmem, > so the range [_text, _stext) is also freed here. Right, I've misread KERNEL_START as if it was _text... > > > > > And we have memory protection mechanism use this > > > which have false alarm when some other IPs doing dma > > > if the dma page frame is in the [_text, _stext). > > > > > > Below are my iomem resource and reserved memory information: > > > console:/ # grep Kernel /proc/iomem > > > 20208000-219fffff : Kernel code > > > 21b00000-21c2e76f : Kernel data > > > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > > 0: 0x20201000..0x20207fff > > > 1: 0x20300000..0x21c2e76f > > > > What are the addresses of _text and _stext in your configuration? > > va pa > c0008000 20208000 _text > c0100000 20300000 _stext > > > > > What these dumps are supposed to show here? > > > > Below is the dump info after applied this patch. > > console:/ # grep Kernel /proc/iomem > 20300000-219fffff : Kernel code > 21b00000-21c2e76f : Kernel data > > console:/ # cat /sys/kernel/debug/memblock/reserved > 0: 0x20201000..0x20207fff > 1: 0x20300000..0x21c2e76f > > The difference is that Kernel Code resource match the reserved memblock 1 > which is reserved in arm_memblock_init(). For that I'd extend the reservation in arm_memblock_init() to include [_text, _stext). Even if the code there is not needed after init, at least we'll keep this consistent with other architectures.
> On Wed, Dec 08, 2021 at 10:26:23AM +0800, Mark-PK Tsai wrote: > > > On Fri, Dec 03, 2021 at 10:14:16PM +0800, Mark-PK Tsai wrote: > > > > Remove the [_text, _stext) from Kernel Code. > > > > Although there are some startup code in head.text, they > > > > are freed to the buddy system after kernel boot. > > > > > > Hmm, I don't see it is being freed anywhere. Can you elaborate when and how > > > the range [_text, _stext) is freed? > > > > arm_memblock_init() reserve [KERNEL_START, KERNEL_END) which are defined as following. > > > > #define KERNEL_START _stext > > #define KERNEL_END _end > > > > free_low_memory_core_early() free all the non-reserved range in lowmem, > > so the range [_text, _stext) is also freed here. > > Right, I've misread KERNEL_START as if it was _text... > > > > > > > > And we have memory protection mechanism use this > > > > which have false alarm when some other IPs doing dma > > > > if the dma page frame is in the [_text, _stext). > > > > > > > > Below are my iomem resource and reserved memory information: > > > > console:/ # grep Kernel /proc/iomem > > > > 20208000-219fffff : Kernel code > > > > 21b00000-21c2e76f : Kernel data > > > > > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > > > 0: 0x20201000..0x20207fff > > > > 1: 0x20300000..0x21c2e76f > > > > > > What are the addresses of _text and _stext in your configuration? > > > > va pa > > c0008000 20208000 _text > > c0100000 20300000 _stext > > > > > > > > What these dumps are supposed to show here? > > > > > > > Below is the dump info after applied this patch. > > > > console:/ # grep Kernel /proc/iomem > > 20300000-219fffff : Kernel code > > 21b00000-21c2e76f : Kernel data > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > 0: 0x20201000..0x20207fff > > 1: 0x20300000..0x21c2e76f > > > > The difference is that Kernel Code resource match the reserved memblock 1 > > which is reserved in arm_memblock_init(). > > For that I'd extend the reservation in arm_memblock_init() to include > [_text, _stext). > > Even if the code there is not needed after init, at least we'll keep this > consistent with other architectures. > Extending the reservation in arm_memblock_init() will consume up to SECTION_SIZE memory for SECTION_SIZE alignment when CONFIG_STRICT_KERNEL_RWX=y. So we really don't want to extend the reservation because it affect the low RAM devices. And arm64_memblock_init() also reserve [_stext, _end) in the recent kernel.
> > On Wed, Dec 08, 2021 at 10:26:23AM +0800, Mark-PK Tsai wrote: > > > > On Fri, Dec 03, 2021 at 10:14:16PM +0800, Mark-PK Tsai wrote: > > > > > Remove the [_text, _stext) from Kernel Code. > > > > > Although there are some startup code in head.text, they > > > > > are freed to the buddy system after kernel boot. > > > > > > > > Hmm, I don't see it is being freed anywhere. Can you elaborate when and how > > > > the range [_text, _stext) is freed? > > > > > > arm_memblock_init() reserve [KERNEL_START, KERNEL_END) which are defined as following. > > > > > > #define KERNEL_START _stext > > > #define KERNEL_END _end > > > > > > free_low_memory_core_early() free all the non-reserved range in lowmem, > > > so the range [_text, _stext) is also freed here. > > > > Right, I've misread KERNEL_START as if it was _text... > > > > > > > > > > > And we have memory protection mechanism use this > > > > > which have false alarm when some other IPs doing dma > > > > > if the dma page frame is in the [_text, _stext). > > > > > > > > > > Below are my iomem resource and reserved memory information: > > > > > console:/ # grep Kernel /proc/iomem > > > > > 20208000-219fffff : Kernel code > > > > > 21b00000-21c2e76f : Kernel data > > > > > > > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > > > > 0: 0x20201000..0x20207fff > > > > > 1: 0x20300000..0x21c2e76f > > > > > > > > What are the addresses of _text and _stext in your configuration? > > > > > > va pa > > > c0008000 20208000 _text > > > c0100000 20300000 _stext > > > > > > > > > > > What these dumps are supposed to show here? > > > > > > > > > > Below is the dump info after applied this patch. > > > > > > console:/ # grep Kernel /proc/iomem > > > 20300000-219fffff : Kernel code > > > 21b00000-21c2e76f : Kernel data > > > > > > console:/ # cat /sys/kernel/debug/memblock/reserved > > > 0: 0x20201000..0x20207fff > > > 1: 0x20300000..0x21c2e76f > > > > > > The difference is that Kernel Code resource match the reserved memblock 1 > > > which is reserved in arm_memblock_init(). > > > > For that I'd extend the reservation in arm_memblock_init() to include > > [_text, _stext). > > > > Even if the code there is not needed after init, at least we'll keep this > > consistent with other architectures. > > > > Extending the reservation in arm_memblock_init() will consume up to SECTION_SIZE > memory for SECTION_SIZE alignment when CONFIG_STRICT_KERNEL_RWX=y. > So we really don't want to extend the reservation because it affect > the low RAM devices. > > And arm64_memblock_init() also reserve [_stext, _end) in the recent kernel. Hi, Is there anything wrong with this patch? Either extend the reserved range or remove the head.text from iomem resource can fix the issue. So if you have any concern about it please let me know. Thanks!
diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c index 284a80c0b6e1..85ffc9501e38 100644 --- a/arch/arm/kernel/setup.c +++ b/arch/arm/kernel/setup.c @@ -851,7 +851,7 @@ static void __init request_standard_resources(const struct machine_desc *mdesc) struct resource *res; u64 i; - kernel_code.start = virt_to_phys(_text); + kernel_code.start = virt_to_phys(_stext); kernel_code.end = virt_to_phys(__init_begin - 1); kernel_data.start = virt_to_phys(_sdata); kernel_data.end = virt_to_phys(_end - 1);
Remove the [_text, _stext) from Kernel Code. Although there are some startup code in head.text, they are freed to the buddy system after kernel boot. And we have memory protection mechanism use this which have false alarm when some other IPs doing dma if the dma page frame is in the [_text, _stext). Below are my iomem resource and reserved memory information: console:/ # grep Kernel /proc/iomem 20208000-219fffff : Kernel code 21b00000-21c2e76f : Kernel data console:/ # cat /sys/kernel/debug/memblock/reserved 0: 0x20201000..0x20207fff 1: 0x20300000..0x21c2e76f Signed-off-by: Mark-PK Tsai <mark-pk.tsai@mediatek.com> --- arch/arm/kernel/setup.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)