From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ch1outboundpool.messaging.microsoft.com (ch1ehsobe006.messaging.microsoft.com [216.32.181.186]) (using TLSv1 with cipher AES128-SHA (128/128 bits)) (Client CN "mail.global.frontbridge.com", Issuer "MSIT Machine Auth CA 2" (not verified)) by ozlabs.org (Postfix) with ESMTPS id E86A82C0302 for ; Wed, 11 Dec 2013 10:19:08 +1100 (EST) Message-ID: <1386717537.10013.150.camel@snotra.buserror.net> Subject: Re: [PATCH v2] powerpc 8xx: Loading kernels over 8Mbytes without CONFIG_PIN_TLB From: Scott Wood To: leroy christophe Date: Tue, 10 Dec 2013 17:18:57 -0600 In-Reply-To: <52A79E31.9070304@c-s.fr> References: <20131210112945.E4E311A2BF3@localhost.localdomain> <1386714253.10013.123.camel@snotra.buserror.net> <52A79E31.9070304@c-s.fr> Content-Type: text/plain; charset="UTF-8" MIME-Version: 1.0 Cc: linuxppc-dev@lists.ozlabs.org, Paul Mackerras , linux-kernel@vger.kernel.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Wed, 2013-12-11 at 00:05 +0100, leroy christophe wrote: > Le 10/12/2013 23:24, Scott Wood a =C3=A9crit : > > On Tue, 2013-12-10 at 12:29 +0100, Christophe Leroy wrote: > >> Today, the only way to load kernels whose size is greater than 8Mbyt= es is to > >> activate CONFIG_PIN_TLB. Otherwise, the physical memory initially ma= pped is > >> limited to 8Mbytes. This patch adds the capability to select the siz= e of initial > >> memory between 8/16/24 Mbytes and this is regardless of whether CONF= IG_PIN_TLB > >> is active or not. It allows to load "big" kernels (for instance when= activating > >> CONFIG_LOCKDEP_SUPPORT) without having to activate CONFIG_PIN_TLB. > >> > >> Signed-off-by: Christophe Leroy > >> > >> diff -ur a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig > >> --- a/arch/powerpc/Kconfig > >> +++ b/arch/powerpc/Kconfig > >> @@ -980,6 +980,29 @@ > >> config PIN_TLB > >> bool "Pinned Kernel TLBs (860 ONLY)" > >> depends on ADVANCED_OPTIONS && 8xx > >> + > >> +choice > >> + prompt "Initial Data Memory Mapped on 8xx" > >> + default 8xx_MAP_8M > >> + depends on ADVANCED_OPTIONS && 8xx > >> + > >> +config 8xx_INIT_MAP_8M > >> + bool "8 Mbytes" > >> + > >> +config 8xx_INIT_MAP_16M > >> + bool "16 Mbytes" > >> + > >> +config 8xx_INIT_MAP_24M > >> + bool "24 Mbytes" > > Are you working with a loader that passes initial-mapped-area size in= r7 > > as per ePAPR? If so, we could rely on that at runtime. If you're us= ing > > a non-ancient U-Boot, it should qualify here even if it's not fully > > ePAPR compliant (it passes the value of the bootm_mapsize variable in > > r7). > Ok, let me check that. But it means that the size of the kernel I can=20 > boot will depend on the initial memory mapped by uboot ? Isn't it=20 > limitating ? The ePAPR IMA is supposed to be large enough to include the OS image, device tree, etc. > Even if uboot only maps 8Mbytes, why couldn't I be allowed to boot a=20 > kernel having 10 Mbytes data if I have 32 Mbytes mem on the board ? > I don't like the idea of having to change the bootloader just because I= =20 > want to activate CONFIG_LOCKDEP to debug my kernel. Well, as noted, if you're using a non-ancient U-Boot you shouldn't have to change anything because it already implements r7. Now, the value of r7 it passes might be a lie as far as ePAPR is concerned, since it's supposed to represent what's actually mapped, but that's another matter. Even fixing that wouldn't mean you have to change U-Boot every time the kernel size changes; you'd just set it to something reasonable and be done with it. I'm not fond of adding kconfigs to hack around a problem that has already been addressed in the standard that governs the PPC boot process that U-Boot claims to implement. > >> -#ifdef CONFIG_PIN_TLB > >> +#if defined (CONFIG_8xx_INIT_MAP_16M) || defined (CONFIG_8xx_INIT_M= AP_24M) > >> /* Map two more 8M kernel data pages. > >> */ > >> +#ifdef CONFIG_PIN_TLB > >> addi r10, r10, 0x0100 > >> mtspr SPRN_MD_CTR, r10 > >> +#endif > >> =20 > >> lis r8, KERNELBASE@h /* Create vaddr for TLB */ > >> addis r8, r8, 0x0080 /* Add 8M */ > >> @@ -858,15 +860,19 @@ > >> addis r11, r11, 0x0080 /* Add 8M */ > >> mtspr SPRN_MD_RPN, r11 > >> =20 > >> +#ifdef CONFIG_8xx_INIT_MAP_24M > >> +#ifdef CONFIG_PIN_TLB > >> addi r10, r10, 0x0100 > >> mtspr SPRN_MD_CTR, r10 > >> +#endif > > Are these ifdefs for CONFIG_PIN_TLB really needed? It shouldn't harm > > anything to use those entries even if they're not being pinned. >=20 > I'm not sure I understand your comment. > ifdef for CONFIG_PIN_TLB was already there before, but was enclosing th= e=20 > whole block, so 24 Mbytes were automatically mapped when you selected=20 > CONFIG_PIN_TLB and only 8 Mbytes were mapped when you didn't select=20 > CONFIG_PIN_TLB. > I reduced the scope of those ifdefs so that they now apply on the=20 > pinning only. There wasn't previously an ifdef specifically around the setting of SPRN_MD_CTR. That's new. There was an ifdef around the entire block, which has gone away because you are now trying to map more than 8M regardless of CONFIG_PIN_TLB, but that has nothing to do with whether there should be an ifdef around SPRN_MD_CTR. -Scott