Merge patch series "Relocate U-Boot in the last bank"

Ilias Apalodimas <[email protected]> says:

There was a discussion recently on the mailing lists regarding our
management of memory above ram_top [0]. The tl;dr is that we have two problems.

The first one is that U-Boot always relocates to the top of the first available
bank unless there's special board code to sidestep that. The second is we don't
successfully deal with devices that can only do 32-bit DMA.

This patch series deals with the first problem by adding a Kconfig option
allowing platforms to relocate to the top of the last discovered bank.

It's worth noting that this is easily testable with QEMU

qemu-system-aarch64 -m 8192 -smp 2 -nographic -cpu cortex-a57 \
    -machine virt,secure=off \
    -bios u-boot.bin \
    -device virtio-rng-pci \
    -drive id=os,if=none,file="$image" \
    -device virtio-blk-device,drive=os \
    -object memory-backend-ram,id=ram0,size=4G \
    -object memory-backend-ram,id=ram1,size=4G \
    -numa node,memdev=ram0 \
    -numa node,memdev=ram1

# RELOC_ADDR_TOP not set
Hit any key to stop autoboot: 0
=> bdinfo
[...]
relocaddr   = 0x000000013f66c000
reloc off   = 0x000000013f66c000
[...]
lmb_dump_all:
 memory.count = 0x1
 memory[0]	[0x40000000-0x23fffffff], 0x200000000 bytes, flags: none
 reserved.count = 0x2
 reserved[0]	[0x13d507000-0x13d509fff], 0x3000 bytes, flags: no-notify, no-overwrite
reserved[1]	[0x13d50aff0-0x23fffffff], 0x102af5010 bytes, flags: no-overwrite
devicetree  = board
[...]
TLB addr    = 0x000000013ffe0000
irq_sp      = 0x000000013e50aff0
sp start    = 0x000000013e50aff0
Early malloc usage: e88 / 2000
=>

# RELOC_ADDR_TOP enabled

=> bdinfo
[...]
relocaddr   = 0x000000023f66c000
reloc off   = 0x000000023f66c000
[...]
lmb_dump_all:
 memory.count = 0x1
 memory[0]	[0x40000000-0x23fffffff], 0x200000000 bytes, flags: none
 reserved.count = 0x2
 reserved[0]	[0x23d507000-0x23d509fff], 0x3000 bytes, flags: no-notify, no-overwrite
 reserved[1]	[0x23d50aff0-0x23fffffff], 0x2af5010 bytes, flags: no-overwrite
devicetree  = board
[...]
TLB addr    = 0x000000023ffe0000
irq_sp      = 0x000000023e50aff0
sp start    = 0x000000023e50aff0
Early malloc usage: e88 / 2000
=>

[0] https://lore.kernel.org/u-boot/CAC_iWjKFAzpj3B_MEW7-dnOrcAV-rfkhXXo8Bv0KgLNP2VJxRA@mail.gmail.com/

Link: https://lore.kernel.org/r/[email protected]

2 files changed, 21 insertions, 21 deletions

diff --git a/lib/fdtdec.c b/lib/fdtdec.c
index d0a84b5034b..b91e067106d 100644
--- a/lib/fdtdec.c
+++ b/lib/fdtdec.c
@@ -35,6 +35,7 @@
 #include <linux/ctype.h>
 #include <linux/lzo.h>
 #include <linux/ioport.h>
+#include <asm/global_data.h>
 
 DECLARE_GLOBAL_DATA_PTR;
 
@@ -1142,14 +1143,14 @@ int fdtdec_setup_memory_banksize(void)
 		if (ret != 0)
 			return -EINVAL;
 
-		gd->bd->bi_dram[bank].start = (phys_addr_t)res.start;
-		gd->bd->bi_dram[bank].size =
+		gd->dram[bank].start = (phys_addr_t)res.start;
+		gd->dram[bank].size =
 			(phys_size_t)(res.end - res.start + 1);
 
 		debug("%s: DRAM Bank #%d: start = %pap, size = %pap\n",
 		      __func__, bank,
-		      &gd->bd->bi_dram[bank].start,
-		      &gd->bd->bi_dram[bank].size);
+		      &gd->dram[bank].start,
+		      &gd->dram[bank].size);
 	}
 
 	return 0;
@@ -1930,7 +1931,7 @@ int fdtdec_resetup(int *rescan)
 
 int fdtdec_decode_ram_size(const void *blob, const char *area, int board_id,
 			   phys_addr_t *basep, phys_size_t *sizep,
-			   struct bd_info *bd)
+			   gd_t *gd_ptr)
 {
 	int addr_cells, size_cells;
 	const u32 *cell, *end;
@@ -1982,8 +1983,8 @@ int fdtdec_decode_ram_size(const void *blob, const char *area, int board_id,
 	}
 	/* Note: if no matching subnode was found we use the parent node */
 
-	if (bd) {
-		memset(bd->bi_dram, '\0', sizeof(bd->bi_dram[0]) *
+	if (gd_ptr) {
+		memset(gd_ptr->dram, '\0', sizeof(gd_ptr->dram[0]) *
 						CONFIG_NR_DRAM_BANKS);
 	}
 
@@ -1999,8 +2000,8 @@ int fdtdec_decode_ram_size(const void *blob, const char *area, int board_id,
 		if (addr_cells == 2)
 			addr += (u64)fdt32_to_cpu(*cell++) << 32UL;
 		addr += fdt32_to_cpu(*cell++);
-		if (bd)
-			bd->bi_dram[bank].start = addr;
+		if (gd_ptr)
+			gd_ptr->dram[bank].start = addr;
 		if (basep && !bank)
 			*basep = (phys_addr_t)addr;
 
@@ -2022,8 +2023,8 @@ int fdtdec_decode_ram_size(const void *blob, const char *area, int board_id,
 			}
 		}
 
-		if (bd)
-			bd->bi_dram[bank].size = size;
+		if (gd_ptr)
+			gd_ptr->dram[bank].size = size;
 		total_size += size;
 	}
 
diff --git a/lib/lmb.c b/lib/lmb.c
index 779df35eb9c..77440a48486 100644
--- a/lib/lmb.c
+++ b/lib/lmb.c
@@ -555,12 +555,12 @@ static void lmb_reserve_uboot_region(void)
 #endif
 
 	for (bank = 0; bank < CONFIG_NR_DRAM_BANKS; bank++) {
-		if (!gd->bd->bi_dram[bank].size ||
-		    rsv_start < gd->bd->bi_dram[bank].start)
+		if (!gd->dram[bank].size ||
+		    rsv_start < gd->dram[bank].start)
 			continue;
 		/* Watch out for RAM at end of address space! */
-		bank_end = gd->bd->bi_dram[bank].start +
-			gd->bd->bi_dram[bank].size - 1;
+		bank_end = gd->dram[bank].start +
+			gd->dram[bank].size - 1;
 		if (rsv_start > bank_end)
 			continue;
 		if (bank_end > end)
@@ -615,7 +615,6 @@ static void lmb_add_memory(void)
 	phys_addr_t bank_end;
 	phys_size_t size;
 	u64 ram_top = gd->ram_top;
-	struct bd_info *bd = gd->bd;
 
 	if (CONFIG_IS_ENABLED(LMB_ARCH_MEM_MAP))
 		return lmb_arch_add_memory();
@@ -625,22 +624,22 @@ static void lmb_add_memory(void)
 		ram_top = 0x100000000ULL;
 
 	for (i = 0; i < CONFIG_NR_DRAM_BANKS; i++) {
-		size = bd->bi_dram[i].size;
+		size = gd->dram[i].size;
 
 		if (size) {
-			lmb_add(bd->bi_dram[i].start, size);
+			lmb_add(gd->dram[i].start, size);
 			if (!IS_ENABLED(CONFIG_LMB_LIMIT_DMA_BELOW_RAM_TOP))
 				continue;
 
-			bank_end = bd->bi_dram[i].start + size;
+			bank_end = gd->dram[i].start + size;
 
 			/*
 			 * Reserve memory above ram_top as
 			 * no-overwrite so that it cannot be
 			 * allocated
 			 */
-			if (bd->bi_dram[i].start >= ram_top)
-				lmb_reserve(bd->bi_dram[i].start, size,
+			if (gd->dram[i].start >= ram_top)
+				lmb_reserve(gd->dram[i].start, size,
 					    LMB_NOOVERWRITE);
 			else if (bank_end > ram_top)
 				lmb_reserve(ram_top, bank_end - ram_top,