Skip to contentSkip to conversation

MIPS: reloc: make flush_cache_all() usable

PATCH In review Antony Pavlov 2 participants 4 messages Jun 21, 2019 – Jul 1, 2019 MIPS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 MIPS: reloc: make flush_cache_all() usable

flush_cache_all() uses 'struct cpuinfo_mips current_cpu_data' data fields. These data fields are initialized in r4k_cache_init().

However in the current implementation the r4k_cache_init() function is called __AFTER__ relocate_code().

---
 arch/mips/lib/reloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/mips/lib/reloc.c +12 -0
index 14ba6167dd..6f1cd6d82f 100644
@@ -37,6 +37,7 @@
 #include <asm/mipsregs.h>
 #include <asm/relocs.h>
 #include <asm/sections.h>
+#include <asm/cpu-features.h>
 #include <linux/sizes.h>
 #include <asm-generic/memory_layout.h>
 
@@ -157,6 +158,17 @@ void relocate_code(void *fdt, u32 fdt_size, u32 ram_size)
 		apply_reloc(type, (void *)addr, off);
 	}
 
+	/* clear the BSS first */
+	memset(__bss_start, 0x00, bss_len);
+
+	cpu_probe();
+
+	if (cpu_has_4k_cache) {
+		extern void r4k_cache_init(void);
+
+		r4k_cache_init();
+	}
+
 	/* Ensure the icache is coherent */
 	flush_cache_all();
 
Signature · 7 lines
2.20.1


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#1 lore raw

Hi Antony,

Quoted 30 lines · Antony Pavlov

flush_cache_all() uses 'struct cpuinfo_mips current_cpu_data' data fields. These data fields are initialized in r4k_cache_init().

However in the current implementation the r4k_cache_init() function is called __AFTER__ relocate_code().

---
 arch/mips/lib/reloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/mips/lib/reloc.c +4 -0
index 14ba6167dd..6f1cd6d82f 100644
@@ -37,6 +37,7 @@
 #include <asm/mipsregs.h>
 #include <asm/relocs.h>
 #include <asm/sections.h>
+#include <asm/cpu-features.h>
 #include <linux/sizes.h>
 #include <asm-generic/memory_layout.h>
 
@@ -157,6 +158,17 @@ void relocate_code(void *fdt, u32 fdt_size, u32 ram_size)
 		apply_reloc(type, (void *)addr, off);
 	}
 
+	/* clear the BSS first */
+	memset(__bss_start, 0x00, bss_len);
+

arch/mips/lib/reloc.c void relocate_code(void *fdt, u32 fdt_size, u32 ram_size) · line 173

Quoted 10 lines
arch/mips/lib/reloc.c +8 -0
+	cpu_probe();
+
+	if (cpu_has_4k_cache) {
+		extern void r4k_cache_init(void);
+
+		r4k_cache_init();
+	}
+
 	/* Ensure the icache is coherent */
 	flush_cache_all();

Can we initialize one field in current_cpu_data to something nonzero so that it goes out of the bss?

Another way would be to pass cpu_probe() a pointer to a struct cpuinfo_mips allocated on the stack and then pass it to flush_cache_all().

Clearing the bss twice doesn't look very nice.

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#2 lore raw
Quoted 30 lines · On Wed, 26 Jun 2019 08:53:54 +0200 Sascha Hauer <[email protected]> wrote:

Hi Antony,

Quoted 27 lines · Antony Pavlov

flush_cache_all() uses 'struct cpuinfo_mips current_cpu_data' data fields. These data fields are initialized in r4k_cache_init().

However in the current implementation the r4k_cache_init() function is called __AFTER__ relocate_code().

---
 arch/mips/lib/reloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/mips/lib/reloc.c +1 -0
index 14ba6167dd..6f1cd6d82f 100644
@@ -37,6 +37,7 @@
 #include <asm/mipsregs.h>
 #include <asm/relocs.h>
 #include <asm/sections.h>
+#include <asm/cpu-features.h>
 #include <linux/sizes.h>
 #include <asm-generic/memory_layout.h>
 
@@ -157,6 +158,17 @@ void relocate_code(void *fdt, u32 fdt_size, u32 ram_size)
 		apply_reloc(type, (void *)addr, off);
 	}
 
Quoted 16 lines

arch/mips/lib/reloc.c void relocate_code(void *fdt, u32 fdt_size, u32 ram_size) · line 173

Quoted 13 lines
arch/mips/lib/reloc.c +11 -0
+	/* clear the BSS first */
+	memset(__bss_start, 0x00, bss_len);
+
+	cpu_probe();
+
+	if (cpu_has_4k_cache) {
+		extern void r4k_cache_init(void);
+
+		r4k_cache_init();
+	}
+
 	/* Ensure the icache is coherent */
 	flush_cache_all();

Can we initialize one field in current_cpu_data to something nonzero so that it goes out of the bss?

Or clear only current_cpu_data and keep it in bss?

Quoted 5 lines

Another way would be to pass cpu_probe() a pointer to a struct cpuinfo_mips allocated on the stack and then pass it to flush_cache_all().

Clearing the bss twice doesn't look very nice.

It's two diffrent bss segments: one before relocation and one after relocation.

Signature · 7 lines
Best regards,
  Antony Pavlov

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#3 lore raw
Quoted 50 lines · Antony Pavlov
Quoted 30 lines · On Wed, 26 Jun 2019 08:53:54 +0200 Sascha Hauer <[email protected]> wrote:

Hi Antony,

Quoted 27 lines · Antony Pavlov

flush_cache_all() uses 'struct cpuinfo_mips current_cpu_data' data fields. These data fields are initialized in r4k_cache_init().

However in the current implementation the r4k_cache_init() function is called __AFTER__ relocate_code().

---
 arch/mips/lib/reloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
arch/mips/lib/reloc.c +1 -0
index 14ba6167dd..6f1cd6d82f 100644
@@ -37,6 +37,7 @@
 #include <asm/mipsregs.h>
 #include <asm/relocs.h>
 #include <asm/sections.h>
+#include <asm/cpu-features.h>
 #include <linux/sizes.h>
 #include <asm-generic/memory_layout.h>
 
@@ -157,6 +158,17 @@ void relocate_code(void *fdt, u32 fdt_size, u32 ram_size)
 		apply_reloc(type, (void *)addr, off);
 	}
 
Quoted 16 lines

arch/mips/lib/reloc.c void relocate_code(void *fdt, u32 fdt_size, u32 ram_size) · line 173

Quoted 13 lines
arch/mips/lib/reloc.c +11 -0
+	/* clear the BSS first */
+	memset(__bss_start, 0x00, bss_len);
+
+	cpu_probe();
+
+	if (cpu_has_4k_cache) {
+		extern void r4k_cache_init(void);
+
+		r4k_cache_init();
+	}
+
 	/* Ensure the icache is coherent */
 	flush_cache_all();

Can we initialize one field in current_cpu_data to something nonzero so that it goes out of the bss?

Quoted 1 line

Or clear only current_cpu_data and keep it in bss?

That would at least make it clear why this is done.

That's important because otherwise some day somebody sends a patch "mips: remove duplicate clearing of bss" and I might apply it because it looks correct ;)

There might be one problem with putting current_cpu_data into bss. I don't know much about MIPS booting, but in case the initial image is located in some potentially small SRAM then putting current_cpu_data into bss means that the SRAM has to be big enough for initial image and bss.

Quoted 8 lines
Quoted 6 lines

Another way would be to pass cpu_probe() a pointer to a struct cpuinfo_mips allocated on the stack and then pass it to flush_cache_all().

Clearing the bss twice doesn't look very nice.

It's two diffrent bss segments: one before relocation and one after relocation.

Yes, I know, otherwise you could just drop the second bss clearing.

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox