[U-Boot-Users] [PATCH] Fix relocation problem with "new" get_dev() function
This patch enables the "new" get_dev() function for block devices introduced by Grant Likely to be used on systems that still suffer from the relocation problems (manual relocation neede because of problems with linker script).
Hopefully we can resolve this relocation issue soon for all platform so we don't need this additional code anymore.
Signed-off-by: Stefan Roese sr@denx.de
--- commit 751bb57107d78978ae08e697c3deba816f5be091 tree 07d9ef94782dadb04203d8f46dd08da5d08a2aa7 parent d93e2212f962668b3dce091ff5edc33f2347fe37 author Stefan Roese sr@denx.de Tue, 20 Feb 2007 13:21:57 +0100 committer Stefan Roese sr@denx.de Tue, 20 Feb 2007 13:21:57 +0100
disk/part.c | 12 ++++++++++++ 1 files changed, 12 insertions(+), 0 deletions(-)
diff --git a/disk/part.c b/disk/part.c index f1026c5..37bba77 100644 --- a/disk/part.c +++ b/disk/part.c @@ -64,13 +64,25 @@ static const struct block_drvr block_drvr[] = { { }, };
+#ifndef CFG_FIXUP_RELOCATION +DECLARE_GLOBAL_DATA_PTR; +#endif + block_dev_desc_t *get_dev(char* ifname, int dev) { const struct block_drvr *drvr = block_drvr;
while (drvr->name) { +#ifndef CFG_FIXUP_RELOCATION + block_dev_desc_t* (*reloc_get_dev)(int dev); + + reloc_get_dev = drvr->get_dev + gd->reloc_off; + if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0) + return reloc_get_dev(dev); +#else if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0) return drvr->get_dev(dev); +#endif drvr++; } return NULL;
On 2/20/07, Stefan Roese sr@denx.de wrote:
This patch enables the "new" get_dev() function for block devices introduced by Grant Likely to be used on systems that still suffer from the relocation problems (manual relocation neede because of problems with linker script).
Hopefully we can resolve this relocation issue soon for all platform so we don't need this additional code anymore.
Signed-off-by: Stefan Roese sr@denx.de
+#ifndef CFG_FIXUP_RELOCATION
block_dev_desc_t* (*reloc_get_dev)(int dev);reloc_get_dev = drvr->get_dev + gd->reloc_off;if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0)return reloc_get_dev(dev);+#else if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0) return drvr->get_dev(dev); +#endif
The changes look good, but I would drop the CFG_FIXUP_RELOCATION macros for now. Just fix it up unconditionally and let the patch that actually fixes relocation add them back in.
I know it doesn't hurt anything to have them in now; but I think the current relocation patch takes the wrong approach. That patch assumes that relocation is busted unless the board config says it isn't. Since this is something that we want to fix correctly; I think it is better to add a '#define CONFIG_RELOCATION_BROKEN' to all board configs files now; and steadly remove them as different ports are fixed.
At first glance this looks like a bad idea because it means every board config file must be touched; but every config file needs to be touched regardless. Either: a) touch every config file now (by adding the "IT'S BROKEN" define; or b) touch every config file later (by removing the "IT'S FIXED" define when all board ports are working and the hooks are no longer needed).
Thoughts? g.
On Wednesday 21 February 2007 18:01, Grant Likely wrote:
On 2/20/07, Stefan Roese sr@denx.de wrote:
This patch enables the "new" get_dev() function for block devices introduced by Grant Likely to be used on systems that still suffer from the relocation problems (manual relocation neede because of problems with linker script).
Hopefully we can resolve this relocation issue soon for all platform so we don't need this additional code anymore.
Signed-off-by: Stefan Roese sr@denx.de
+#ifndef CFG_FIXUP_RELOCATION
block_dev_desc_t* (*reloc_get_dev)(int dev);reloc_get_dev = drvr->get_dev + gd->reloc_off;if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0)return reloc_get_dev(dev);+#else if (strncmp(ifname, drvr->name, strlen(drvr->name)) == 0) return drvr->get_dev(dev); +#endif
The changes look good, but I would drop the CFG_FIXUP_RELOCATION macros for now. Just fix it up unconditionally and let the patch that actually fixes relocation add them back in.
Yes, that does make more sense for now. I'll update my version.
I know it doesn't hurt anything to have them in now; but I think the current relocation patch takes the wrong approach. That patch assumes that relocation is busted unless the board config says it isn't. Since this is something that we want to fix correctly; I think it is better to add a '#define CONFIG_RELOCATION_BROKEN' to all board configs files now; and steadly remove them as different ports are fixed.
At first glance this looks like a bad idea because it means every board config file must be touched; but every config file needs to be touched regardless. Either: a) touch every config file now (by adding the "IT'S BROKEN" define; or b) touch every config file later (by removing the "IT'S FIXED" define when all board ports are working and the hooks are no longer needed).
Thoughts?
I like this approach, since it seems to be the "right way" to fix this problem. Any other ideas on this? Please speak up, since it's quite a big change.
Best regards, Stefan
===================================================================== DENX Software Engineering GmbH, HRB 165235 Munich, CEO: Wolfgang Denk Office: Kirchenstr. 5, D-82194 Groebenzell, Germany =====================================================================
On 2/21/07, Stefan Roese sr@denx.de wrote:
On Wednesday 21 February 2007 18:01, Grant Likely wrote:
Since this is something that we want to fix correctly; I think it is better to add a '#define CONFIG_RELOCATION_BROKEN' to all board configs files now; and steadly remove them as different ports are fixed.
Thoughts?
I like this approach, since it seems to be the "right way" to fix this problem. Any other ideas on this? Please speak up, since it's quite a big change.
I'll compose my reply in patch form... :)
g.
participants (2)
-
Grant Likely -
Stefan Roese