| Message ID | 20220506114402.365-1-thunder.leizhen@huawei.com |
|---|---|
| Headers | show |
| Series | support reserving crashkernel above 4G on arm64 kdump | expand |
On 05/06/22 at 07:43pm, Zhen Lei wrote: ...... > @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) > if (crash_base) > crash_max = crash_base + crash_size; > > - /* Current arm64 boot protocol requires 2MB alignment */ > - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, > + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, > crash_base, crash_max); > if (!crash_base) { > pr_warn("cannot allocate crashkernel (size:0x%llx)\n", > @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) > return; > } > There's corner case missed, e.g 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; 2) ,high and ,low are specified, the whole system memory is under 4G. Below judgement can filter them away: if (crash_base > arm64_dma_phys_limit && crash_low_size && reserve_crashkernel_low(crash_low_size)) { What's your opinion? Leave it and add document to notice user, or fix it with code change? I would suggest merging this series, Lei can add this corner case handling on top. Since this is a newly added support, we don't have to make it one step. Doing step by step can make reviewing easier. > + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { > + memblock_phys_free(crash_base, crash_size); > + return; > + } > + > pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", > crash_base, crash_base + crash_size, crash_size >> 20); > > @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) > * map. Inform kmemleak so that it won't try to access it. > */ > kmemleak_ignore_phys(crash_base); > + if (crashk_low_res.end) > + kmemleak_ignore_phys(crashk_low_res.start); > + > crashk_res.start = crash_base; > crashk_res.end = crash_base + crash_size - 1; > insert_resource(&iomem_resource, &crashk_res); > -- > 2.25.1 >
On 05/06/22 at 07:44pm, Zhen Lei wrote: > From: Chen Zhou <chenzhou10@huawei.com> > > When reserving crashkernel in high memory, some low memory is reserved > for crash dump kernel devices and never mapped by the first kernel. > This memory range is advertised to crash dump kernel via DT property > under /chosen, > linux,usable-memory-range = <BASE1 SIZE1 [BASE2 SIZE2]> > > We reused the DT property linux,usable-memory-range and made the low > memory region as the second range "BASE2 SIZE2", which keeps compatibility > with existing user-space and older kdump kernels. > > Crash dump kernel reads this property at boot time and call memblock_add() > to add the low memory region after memblock_cap_memory_range() has been > called. LGTM, Acked-by: Baoquan He <bhe@redhat.com> > > Signed-off-by: Chen Zhou <chenzhou10@huawei.com> > Co-developed-by: Zhen Lei <thunder.leizhen@huawei.com> > Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> > Reviewed-by: Rob Herring <robh@kernel.org> > Tested-by: Dave Kleikamp <dave.kleikamp@oracle.com> > --- > drivers/of/fdt.c | 33 +++++++++++++++++++++++---------- > 1 file changed, 23 insertions(+), 10 deletions(-) > > diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c > index ec315b060cd50d2..2f248d0acc04830 100644 > --- a/drivers/of/fdt.c > +++ b/drivers/of/fdt.c > @@ -973,16 +973,24 @@ static void __init early_init_dt_check_for_elfcorehdr(unsigned long node) > > static unsigned long chosen_node_offset = -FDT_ERR_NOTFOUND; > > +/* > + * The main usage of linux,usable-memory-range is for crash dump kernel. > + * Originally, the number of usable-memory regions is one. Now there may > + * be two regions, low region and high region. > + * To make compatibility with existing user-space and older kdump, the low > + * region is always the last range of linux,usable-memory-range if exist. > + */ > +#define MAX_USABLE_RANGES 2 > + > /** > * early_init_dt_check_for_usable_mem_range - Decode usable memory range > * location from flat tree > */ > void __init early_init_dt_check_for_usable_mem_range(void) > { > - const __be32 *prop; > - int len; > - phys_addr_t cap_mem_addr; > - phys_addr_t cap_mem_size; > + struct memblock_region rgn[MAX_USABLE_RANGES] = {0}; > + const __be32 *prop, *endp; > + int len, i; > unsigned long node = chosen_node_offset; > > if ((long)node < 0) > @@ -991,16 +999,21 @@ void __init early_init_dt_check_for_usable_mem_range(void) > pr_debug("Looking for usable-memory-range property... "); > > prop = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len); > - if (!prop || (len < (dt_root_addr_cells + dt_root_size_cells))) > + if (!prop || (len % (dt_root_addr_cells + dt_root_size_cells))) > return; > > - cap_mem_addr = dt_mem_next_cell(dt_root_addr_cells, &prop); > - cap_mem_size = dt_mem_next_cell(dt_root_size_cells, &prop); > + endp = prop + (len / sizeof(__be32)); > + for (i = 0; i < MAX_USABLE_RANGES && prop < endp; i++) { > + rgn[i].base = dt_mem_next_cell(dt_root_addr_cells, &prop); > + rgn[i].size = dt_mem_next_cell(dt_root_size_cells, &prop); > > - pr_debug("cap_mem_start=%pa cap_mem_size=%pa\n", &cap_mem_addr, > - &cap_mem_size); > + pr_debug("cap_mem_regions[%d]: base=%pa, size=%pa\n", > + i, &rgn[i].base, &rgn[i].size); > + } > > - memblock_cap_memory_range(cap_mem_addr, cap_mem_size); > + memblock_cap_memory_range(rgn[0].base, rgn[0].size); > + for (i = 1; i < MAX_USABLE_RANGES && rgn[i].size; i++) > + memblock_add(rgn[i].base, rgn[i].size); > } > > #ifdef CONFIG_SERIAL_EARLYCON > -- > 2.25.1 >
On 2022/5/7 7:10, Baoquan He wrote: > On 05/06/22 at 07:43pm, Zhen Lei wrote: > ...... >> @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) >> if (crash_base) >> crash_max = crash_base + crash_size; >> >> - /* Current arm64 boot protocol requires 2MB alignment */ >> - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, >> + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, >> crash_base, crash_max); >> if (!crash_base) { >> pr_warn("cannot allocate crashkernel (size:0x%llx)\n", >> @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) >> return; >> } >> > > There's corner case missed, e.g > 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; > 2) ,high and ,low are specified, the whole system memory is under 4G. > > Below judgement can filter them away: > > if (crash_base > arm64_dma_phys_limit && crash_low_size && > reserve_crashkernel_low(crash_low_size)) { > > What's your opinion? Leave it and add document to notice user, or fix it > with code change? I think maybe we can leave it unchanged. If the user configures two memory ranges, we'd better apply for two. Otherwise, he'll be confused when he inquires. Currently, crash_low_size is non-zero only when 'crashkernel=Y,low' is explicitly configured. > > I would suggest merging this series, Lei can add this corner case > handling on top. Since this is a newly added support, we don't have > to make it one step. Doing step by step can make reviewing easier. > >> + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { >> + memblock_phys_free(crash_base, crash_size); >> + return; >> + } >> + >> pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", >> crash_base, crash_base + crash_size, crash_size >> 20); >> >> @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) >> * map. Inform kmemleak so that it won't try to access it. >> */ >> kmemleak_ignore_phys(crash_base); >> + if (crashk_low_res.end) >> + kmemleak_ignore_phys(crashk_low_res.start); >> + >> crashk_res.start = crash_base; >> crashk_res.end = crash_base + crash_size - 1; >> insert_resource(&iomem_resource, &crashk_res); >> -- >> 2.25.1 >> > > . >
On 2022/5/7 10:07, Baoquan He wrote: > On 05/07/22 at 09:34am, Leizhen (ThunderTown) wrote: >> >> >> On 2022/5/7 7:10, Baoquan He wrote: >>> On 05/06/22 at 07:43pm, Zhen Lei wrote: >>> ...... >>>> @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) >>>> if (crash_base) >>>> crash_max = crash_base + crash_size; >>>> >>>> - /* Current arm64 boot protocol requires 2MB alignment */ >>>> - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, >>>> + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, >>>> crash_base, crash_max); >>>> if (!crash_base) { >>>> pr_warn("cannot allocate crashkernel (size:0x%llx)\n", >>>> @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) >>>> return; >>>> } >>>> >>> >>> There's corner case missed, e.g >>> 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; >>> 2) ,high and ,low are specified, the whole system memory is under 4G. >>> >>> Below judgement can filter them away: >>> >>> if (crash_base > arm64_dma_phys_limit && crash_low_size && >>> reserve_crashkernel_low(crash_low_size)) { >>> >>> What's your opinion? Leave it and add document to notice user, or fix it >>> with code change? >> >> I think maybe we can leave it unchanged. If the user configures two memory ranges, >> we'd better apply for two. Otherwise, he'll be confused when he inquires. Currently, >> crash_low_size is non-zero only when 'crashkernel=Y,low' is explicitly configured. > > Then user need know the system information, e.g how much is the high > memory, low memory, if CONFIG_ZONE_DMA|DMA32 is enabled. And we need > describe these cases in document. Any corner case or exception need > be noted if we don't handle it in code. > > Caring about this very much because we have CI with existed test cases > to run on the system, and QA will check these manually too. Support > engineer need detailed document if anything special but happened. > Anything unclear or uncovered will be reported as bug to our kernel dev. > Guess your company do the similar thing like this. > > This crashkerne,high and crashkernel,low reservation is special if we > allow ,high, ,low existing in the same zone. Imagine on system with > CONFIG_ZONE_DMA|DMA32 disabled, people copy the crashkernel=512M,high > and crashkernel=128M,low from other system, and he could get > crash_res at [5G, 5G+512M], while crash_low_res at [6G, 6G+128M]. Guess > how they will judge us. OK, I got it. > >> >>> >>> I would suggest merging this series, Lei can add this corner case >>> handling on top. Since this is a newly added support, we don't have >>> to make it one step. Doing step by step can make reviewing easier. >>> >>>> + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { >>>> + memblock_phys_free(crash_base, crash_size); >>>> + return; >>>> + } >>>> + >>>> pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", >>>> crash_base, crash_base + crash_size, crash_size >> 20); >>>> >>>> @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) >>>> * map. Inform kmemleak so that it won't try to access it. >>>> */ >>>> kmemleak_ignore_phys(crash_base); >>>> + if (crashk_low_res.end) >>>> + kmemleak_ignore_phys(crashk_low_res.start); >>>> + >>>> crashk_res.start = crash_base; >>>> crashk_res.end = crash_base + crash_size - 1; >>>> insert_resource(&iomem_resource, &crashk_res); >>>> -- >>>> 2.25.1 >>>> >>> >>> . >>> >> >> -- >> Regards, >> Zhen Lei >> > > . >
On 2022/5/7 11:37, Leizhen (ThunderTown) wrote: > > > On 2022/5/7 10:07, Baoquan He wrote: >> On 05/07/22 at 09:34am, Leizhen (ThunderTown) wrote: >>> >>> >>> On 2022/5/7 7:10, Baoquan He wrote: >>>> On 05/06/22 at 07:43pm, Zhen Lei wrote: >>>> ...... >>>>> @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) >>>>> if (crash_base) >>>>> crash_max = crash_base + crash_size; >>>>> >>>>> - /* Current arm64 boot protocol requires 2MB alignment */ >>>>> - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, >>>>> + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, >>>>> crash_base, crash_max); >>>>> if (!crash_base) { >>>>> pr_warn("cannot allocate crashkernel (size:0x%llx)\n", >>>>> @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) >>>>> return; >>>>> } >>>>> >>>> >>>> There's corner case missed, e.g >>>> 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; >>>> 2) ,high and ,low are specified, the whole system memory is under 4G. >>>> >>>> Below judgement can filter them away: >>>> >>>> if (crash_base > arm64_dma_phys_limit && crash_low_size && >>>> reserve_crashkernel_low(crash_low_size)) { >>>> >>>> What's your opinion? Leave it and add document to notice user, or fix it >>>> with code change? I decided to modify the code and document. But the code changes aren't what you suggested. For the following reasons: 1. The memory allocated for 'high' may be partially under 4G. So the low memory may not be enough. Of course, it's rare. 2. The second kernel can work properly only when the high and low memory are successfully applied for. For example, high=128M, low=128M, but the second kernel need 256M. So for the cases you listed: 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; --> Follow you suggestion, ignore crashkernel=Y,low, don't allocate low memory. @@ -100,6 +100,14 @@ static int __init reserve_crashkernel_low(unsigned long long low_size) { unsigned long long low_base; + /* + * The kernel does not have any DMA zone, so the range of each DMA + * zone is unknown. Please make sure both CONFIG_ZONE_DMA and + * CONFIG_ZONE_DMA32 are also not set in the second kernel. + */ + if (!IS_ENABLED(CONFIG_ZONE_DMA) && !IS_ENABLED(CONFIG_ZONE_DMA32)) + return 0; + 2) ,high and ,low are specified, the whole system memory is under 4G. --> two memory ranges will be allocated, the size is what 'high' and 'low' specified. --> Yes, the memory of 'low' may be above 'high', but the 'high' just hint allocation --> from top, try high memory first. Of course, this may cause kexec to fail to load. --> Because the memory of 'low' with small size will be used to store Image, etc.. --> But the memory of 'low' above 'high' is almost impossible, we use memblock API to --> allocate memory from top to bottem, 'low' above 'high' need a sizeable memory block --> (128M, 256M?) to be freed at init phase. --> Maybe I should add: crash_max = min(crash_base, CRASH_ADDR_LOW_MAX); --> to make sure the memory of 'low' is always under 'high' >>> >>> I think maybe we can leave it unchanged. If the user configures two memory ranges, >>> we'd better apply for two. Otherwise, he'll be confused when he inquires. Currently, >>> crash_low_size is non-zero only when 'crashkernel=Y,low' is explicitly configured. >> >> Then user need know the system information, e.g how much is the high >> memory, low memory, if CONFIG_ZONE_DMA|DMA32 is enabled. And we need >> describe these cases in document. Any corner case or exception need >> be noted if we don't handle it in code. >> >> Caring about this very much because we have CI with existed test cases >> to run on the system, and QA will check these manually too. Support >> engineer need detailed document if anything special but happened. >> Anything unclear or uncovered will be reported as bug to our kernel dev. >> Guess your company do the similar thing like this. >> >> This crashkerne,high and crashkernel,low reservation is special if we >> allow ,high, ,low existing in the same zone. Imagine on system with >> CONFIG_ZONE_DMA|DMA32 disabled, people copy the crashkernel=512M,high >> and crashkernel=128M,low from other system, and he could get >> crash_res at [5G, 5G+512M], while crash_low_res at [6G, 6G+128M]. Guess >> how they will judge us. > > OK, I got it. > >> >>> >>>> >>>> I would suggest merging this series, Lei can add this corner case >>>> handling on top. Since this is a newly added support, we don't have >>>> to make it one step. Doing step by step can make reviewing easier. >>>> >>>>> + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { >>>>> + memblock_phys_free(crash_base, crash_size); >>>>> + return; >>>>> + } >>>>> + >>>>> pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", >>>>> crash_base, crash_base + crash_size, crash_size >> 20); >>>>> >>>>> @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) >>>>> * map. Inform kmemleak so that it won't try to access it. >>>>> */ >>>>> kmemleak_ignore_phys(crash_base); >>>>> + if (crashk_low_res.end) >>>>> + kmemleak_ignore_phys(crashk_low_res.start); >>>>> + >>>>> crashk_res.start = crash_base; >>>>> crashk_res.end = crash_base + crash_size - 1; >>>>> insert_resource(&iomem_resource, &crashk_res); >>>>> -- >>>>> 2.25.1 >>>>> >>>> >>>> . >>>> >>> >>> -- >>> Regards, >>> Zhen Lei >>> >> >> . >> >
On 05/07/22 at 05:35pm, Leizhen (ThunderTown) wrote: > > > On 2022/5/7 11:37, Leizhen (ThunderTown) wrote: > > > > > > On 2022/5/7 10:07, Baoquan He wrote: > >> On 05/07/22 at 09:34am, Leizhen (ThunderTown) wrote: > >>> > >>> > >>> On 2022/5/7 7:10, Baoquan He wrote: > >>>> On 05/06/22 at 07:43pm, Zhen Lei wrote: > >>>> ...... > >>>>> @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) > >>>>> if (crash_base) > >>>>> crash_max = crash_base + crash_size; > >>>>> > >>>>> - /* Current arm64 boot protocol requires 2MB alignment */ > >>>>> - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, > >>>>> + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, > >>>>> crash_base, crash_max); > >>>>> if (!crash_base) { > >>>>> pr_warn("cannot allocate crashkernel (size:0x%llx)\n", > >>>>> @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) > >>>>> return; > >>>>> } > >>>>> > >>>> > >>>> There's corner case missed, e.g > >>>> 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; > >>>> 2) ,high and ,low are specified, the whole system memory is under 4G. > >>>> > >>>> Below judgement can filter them away: > >>>> > >>>> if (crash_base > arm64_dma_phys_limit && crash_low_size && > >>>> reserve_crashkernel_low(crash_low_size)) { > >>>> > >>>> What's your opinion? Leave it and add document to notice user, or fix it > >>>> with code change? > > I decided to modify the code and document. But the code changes aren't what > you suggested. For the following reasons: Hi Lei, I would say let's merge this version firstly, then add the left step by step. Crashkernel= is not a simple parameter, expecting to make it in one step is not realistic. Otherwise, we will be in a mess of all cases of discussion and handling. Let's slow down and get the basic support added. > 1. The memory allocated for 'high' may be partially under 4G. So the low > memory may not be enough. Of course, it's rare. No, let's forget under 4G or above 4G thing on arm64, but use arm64_dma_phys_limit instead. It's basically equivalent to 4G on x86, while will cause confusion. And I may not get what you are saying the 'high' partially under 4G thing, could you be more specific or give an example? > 2. The second kernel can work properly only when the high and low memory > are successfully applied for. For example, high=128M, low=128M, but the > second kernel need 256M. I may not get this either. We usually won't split our memory requirement into ,high and ,low region. ,high is the main place to accommadate kernel image, initrd, and user space program's memory allocation. ,low is for DMA during kernel bootup. We probably should not encourage or guide user to use like this if I got you correctly. That will complicate the crashkernel= usage more. > > So for the cases you listed: > 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; > --> Follow you suggestion, ignore crashkernel=Y,low, don't allocate low memory. > > @@ -100,6 +100,14 @@ static int __init reserve_crashkernel_low(unsigned long long low_size) > { > unsigned long long low_base; > > + /* > + * The kernel does not have any DMA zone, so the range of each DMA > + * zone is unknown. Please make sure both CONFIG_ZONE_DMA and > + * CONFIG_ZONE_DMA32 are also not set in the second kernel. > + */ > + if (!IS_ENABLED(CONFIG_ZONE_DMA) && !IS_ENABLED(CONFIG_ZONE_DMA32)) > + return 0; > + > > 2) ,high and ,low are specified, the whole system memory is under 4G. > --> two memory ranges will be allocated, the size is what 'high' and 'low' specified. > --> Yes, the memory of 'low' may be above 'high', but the 'high' just hint allocation > --> from top, try high memory first. Of course, this may cause kexec to fail to load. > --> Because the memory of 'low' with small size will be used to store Image, etc.. > --> But the memory of 'low' above 'high' is almost impossible, we use memblock API to > --> allocate memory from top to bottem, 'low' above 'high' need a sizeable memory block > --> (128M, 256M?) to be freed at init phase. Not really. Please think about the case that crashkernel=1G,hign crashkernel=128M,low. memblock top down allocation find a lower position for 1G, but a higher position for 128M because of meomry fragmentation. It's easy but reasonable thing. > --> Maybe I should add: crash_max = min(crash_base, CRASH_ADDR_LOW_MAX); > --> to make sure the memory of 'low' is always under 'high' I would say let's not scatter these details into different places. Like what I changed, it's much easier and code is more understandable. Let's discuss this after this series accepted. A new series can be posted to handle these. > > >>> > >>> I think maybe we can leave it unchanged. If the user configures two memory ranges, > >>> we'd better apply for two. Otherwise, he'll be confused when he inquires. Currently, > >>> crash_low_size is non-zero only when 'crashkernel=Y,low' is explicitly configured. > >> > >> Then user need know the system information, e.g how much is the high > >> memory, low memory, if CONFIG_ZONE_DMA|DMA32 is enabled. And we need > >> describe these cases in document. Any corner case or exception need > >> be noted if we don't handle it in code. > >> > >> Caring about this very much because we have CI with existed test cases > >> to run on the system, and QA will check these manually too. Support > >> engineer need detailed document if anything special but happened. > >> Anything unclear or uncovered will be reported as bug to our kernel dev. > >> Guess your company do the similar thing like this. > >> > >> This crashkerne,high and crashkernel,low reservation is special if we > >> allow ,high, ,low existing in the same zone. Imagine on system with > >> CONFIG_ZONE_DMA|DMA32 disabled, people copy the crashkernel=512M,high > >> and crashkernel=128M,low from other system, and he could get > >> crash_res at [5G, 5G+512M], while crash_low_res at [6G, 6G+128M]. Guess > >> how they will judge us. > > > > OK, I got it. > > > >> > >>> > >>>> > >>>> I would suggest merging this series, Lei can add this corner case > >>>> handling on top. Since this is a newly added support, we don't have > >>>> to make it one step. Doing step by step can make reviewing easier. > >>>> > >>>>> + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { > >>>>> + memblock_phys_free(crash_base, crash_size); > >>>>> + return; > >>>>> + } > >>>>> + > >>>>> pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", > >>>>> crash_base, crash_base + crash_size, crash_size >> 20); > >>>>> > >>>>> @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) > >>>>> * map. Inform kmemleak so that it won't try to access it. > >>>>> */ > >>>>> kmemleak_ignore_phys(crash_base); > >>>>> + if (crashk_low_res.end) > >>>>> + kmemleak_ignore_phys(crashk_low_res.start); > >>>>> + > >>>>> crashk_res.start = crash_base; > >>>>> crashk_res.end = crash_base + crash_size - 1; > >>>>> insert_resource(&iomem_resource, &crashk_res); > >>>>> -- > >>>>> 2.25.1 > >>>>> > >>>> > >>>> . > >>>> > >>> > >>> -- > >>> Regards, > >>> Zhen Lei > >>> > >> > >> . > >> > > > > -- > Regards, > Zhen Lei >
On 5/6/22 6:10 PM, Baoquan He wrote: > On 05/06/22 at 07:43pm, Zhen Lei wrote: > ...... >> @@ -118,8 +162,7 @@ static void __init reserve_crashkernel(void) >> if (crash_base) >> crash_max = crash_base + crash_size; >> >> - /* Current arm64 boot protocol requires 2MB alignment */ >> - crash_base = memblock_phys_alloc_range(crash_size, SZ_2M, >> + crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN, >> crash_base, crash_max); >> if (!crash_base) { >> pr_warn("cannot allocate crashkernel (size:0x%llx)\n", >> @@ -127,6 +170,11 @@ static void __init reserve_crashkernel(void) >> return; >> } >> > > There's corner case missed, e.g > 1) ,high and ,low are specified, CONFIG_ZONE_DMA|DMA32 is not enabled; > 2) ,high and ,low are specified, the whole system memory is under 4G. > > Below judgement can filter them away: > > if (crash_base > arm64_dma_phys_limit && crash_low_size && > reserve_crashkernel_low(crash_low_size)) { > > What's your opinion? Leave it and add document to notice user, or fix it > with code change > > I would suggest merging this series, Lei can add this corner case > handling on top. Since this is a newly added support, we don't have > to make it one step. Doing step by step can make reviewing easier. Lets get this added and tested with a broader audience. It has been in review since March, 4th 2019 - 3+ years. I applaud Zhen for his endurance and patience in carrying this for so long. > >> + if (crash_low_size && reserve_crashkernel_low(crash_low_size)) { >> + memblock_phys_free(crash_base, crash_size); >> + return; >> + } >> + >> pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", >> crash_base, crash_base + crash_size, crash_size >> 20); >> >> @@ -135,6 +183,9 @@ static void __init reserve_crashkernel(void) >> * map. Inform kmemleak so that it won't try to access it. >> */ >> kmemleak_ignore_phys(crash_base); >> + if (crashk_low_res.end) >> + kmemleak_ignore_phys(crashk_low_res.start); >> + >> crashk_res.start = crash_base; >> crashk_res.end = crash_base + crash_size - 1; >> insert_resource(&iomem_resource, &crashk_res); >> -- >> 2.25.1 >> >
On Fri, 6 May 2022 19:43:56 +0800, Zhen Lei wrote: > Changes since [v23]: > Do two minor updates: > 1. Change the value of CRASH_ADDR_HIGH_MAX from memblock.current_limit to (PHYS_MASK + 1). > 2. The 'high' and 'low' are only take effect when crashkernel=X[@offset] does not exist. > If the value of crashkernel=X is invalid, the reservation of crash fails, even if 'high' > and 'low' are configured correctly. > > [...] Applied to arm64 (for-next/crashkernel), finally ;). Thanks! Only patches on top please for the fall-back implementation. [1/6] kdump: return -ENOENT if required cmdline option does not exist https://git.kernel.org/arm64/c/2e5920bb073a [2/6] arm64: Use insert_resource() to simplify code https://git.kernel.org/arm64/c/e6b394425c61 [3/6] arm64: kdump: Reimplement crashkernel=X https://git.kernel.org/arm64/c/944a45abfabc [4/6] of: fdt: Add memory for devices by DT property "linux,usable-memory-range" https://git.kernel.org/arm64/c/fb319e77a0e7 [5/6] of: Support more than one crash kernel regions for kexec -s https://git.kernel.org/arm64/c/8af6b91f5834 [6/6] docs: kdump: Update the crashkernel description for arm64 https://git.kernel.org/arm64/c/5832f1ae5060
Changes since [v23]: Do two minor updates: 1. Change the value of CRASH_ADDR_HIGH_MAX from memblock.current_limit to (PHYS_MASK + 1). 2. The 'high' and 'low' are only take effect when crashkernel=X[@offset] does not exist. If the value of crashkernel=X is invalid, the reservation of crash fails, even if 'high' and 'low' are configured correctly. The test cases are also updated: 1) crashkernel=4G //high=4G, failed 2) crashkernel=256M crashkernel=512M,high crashkernel=512M,low //low=256M, high and low are ignored 3) crashkernel=256M crashkernel=512M,high //low=256M, high is ignored 4) crashkernel=256M crashkernel=512M,low //low=256M, low is ignored 5) crashkernel=1G@0xe0000000 //high=0G, low=0M, cannot allocate, failed 6) crashkernel=512M //high=0G, low=512M 7) crashkernel=128M //high=0G, low=128M 8) crashkernel=512M@0xde000000 //512M@3552M //high=0G, low=512M 9) crashkernel=4G,high //high=4G, low=0M a) crashkernel=4G,high crashkernel=512M,low //high=4G, low=512M b) crashkernel=512M,high crashkernel=128M,low //high=512M, low=128M c) crashkernel=128M,high //high=128M, low=0M d) crashkernel=512M,low //high=0G, low=0M, invalid e) crashkernel=512M,high crashkernel=0,low //high=512M, low=0M f) crashkernel=4G,high crashkernel=ab,low //high=0G, low=0M, invalid g) crashkernel=0M crashkernel=512M,high //high=0G, low=0M, invalid h) crashkernel=abM crashkernel=512M,high //high=0G, low=0M, invalid Changes since [v22]: 1. Add patch 5/6 to support kexec option -s when 'high' and 'low' are used. 2. Remove NO_BLOCK_MAPPINGS optimisations. 3. Don't fallback to high memory when crashkernel=X fails to reserve memory from DMA zone. 4. If "crashkernel=X,low" is not specified, no extra DMA zone memory will be allocated, equivalent to "crashkernel=0,low". 5. The doc has been modified accordingly, because of 3,4 above. Changes since [v21]: 1. Update the commit message of patch 1 and 5. 2. Add some comments for reserve_crashkernel() in patch 5. Thanks to Baoquan He and John Donnelly for their review comments. Because v5.18-rc1 has added a new patch commit 031495635b46 ("arm64: Do not defer reserve_crashkernel() for platforms with no DMA memory zones") There are many new scenarios: 1) The mappings may be block or page-level. 2) The call to reserve_crashkernel() may or may not be deferred. 3) The the upper limit of DMA address may be 4G, or less than 4G. Or the upper limit of physical memory, because SMMU can do the mapping. The code of patch 1-2, 8-9 keep no change, because the above-mentioned issues are not involved. The code of patch 5 only makes the following changes: - if (crash_base >= SZ_4G) + /* + * When both CONFIG_ZONE_DMA and CONFIG_ZONE_DMA32 are disabled, the + * CRASH_ADDR_LOW_MAX equals the upper limit of physical memory, so + * the 'crash_base' of high memory can not exceed it. To follow the + * description of "crashkernel=X,high" option, add below 'high' + * condition to make sure the crash low memory will be reserved. + */ + if ((crash_base >= CRASH_ADDR_LOW_MAX) || high) { Change SZ_4G to CRASH_ADDR_LOW_MAX, because arm64_dma_phys_limit may be less than 4G or greater than 4G. The check 'high' is used for "crashkernel=X,high" and "(crash_base >= CRASH_ADDR_LOW_MAX)" is used for "crashkernel=X[@offset]". Patch 3-4 to allow block mappings for memory above 4G. Patch 6-7 to support only crash high memory or fixed memory range specified by crashkernel=X@offset use page-level mapping, to allow other areas use block mapping. These four patches are for performance optimization purposes. For details about the technical feasibility analysis, please see the commit messages. Now the implementation of arm64 is very different from that of x86. It's no longer suitable for both of them to share code. Changes since [v20]: 1. Check whether crashkernel=Y,low is incorrectly configured or not configured. Do different processing. 2. Share the existing description of x86. The configuration of arm64 is the same as that of x86. 3. Define the value of macro CRASH_ADDR_HIGH_MAX as memblock.current_limit, instead of MEMBLOCK_ALLOC_ACCESSIBLE. 4. To improve readability, some lightweight code adjustments have been made to reserve_craskernel(), including comments. 5. The defined value of DEFAULT_CRASH_KERNEL_LOW_SIZE reconsiders swiotlb, just like x86, to share documents. Thanks to Baoquan He for his careful review. The test cases are as follows: (Please update the kexec tool to the latest version) 1) crashkernel=4G //high=4G, low=256M 2) crashkernel=4G crashkernel=512M,high crashkernel=512M,low //high=4G, low=256M, high and low are ignored 3) crashkernel=4G crashkernel=512M,high //high=4G, low=256M, high is ignored 4) crashkernel=4G crashkernel=512M,low //high=4G, low=256M, low is ignored 5) crashkernel=4G@0xe0000000 //high=0G, low=0M, cannot allocate, failed 6) crashkernel=512M //high=0G, low=512M 7) crashkernel=128M //high=0G, low=128M 8) crashkernel=512M@0xde000000 //512M@3552M //high=0G, low=512M 9) crashkernel=4G,high //high=4G, low=256M a) crashkernel=4G,high crashkernel=512M,low //high=4G, low=512M b) crashkernel=512M,high crashkernel=128M,low //high=512M, low=128M c) crashkernel=128M,high //high=128M, low=256M d) crashkernel=512M,low //high=0G, low=0M, invalid e) crashkernel=512M,high crashkernel=0,low //high=512M, low=0M f) crashkernel=4G,high crashkernel=ab,low //high=0G, low=0M, invalid Changes since [v19]: 1. Temporarily stop making reserve_crashkernel[_low]() generic. There are a lot of details need to be considered, which can take a long time. Because "make generic" does not add new functions and does not improve performance, maybe I should say it's just a cleanup. So by stripping it out and leaving it for other patches later, we can aggregate the changes to the main functions. 2. Use insert_resource() to replace request_resource(), this not only simplifies the code, but also reduces the differences between arm64 and x86 implementations. 3. As commit 157752d84f5d ("kexec: use Crash kernel for Crash kernel low") do for x86, we can also extend kexec-tools for arm64, and it's currently applied. See: https://www.spinics.net/lists/kexec/msg28284.html Thank you very much, Borislav Petkov, for so many valuable comments. Changes since [v17]: v17 --> v19 1. Patch 0001-0004 Introduce generic parse_crashkernel_high_low() to bring the parsing of "crashkernel=X,high" and the parsing of "crashkernel=X,low" together, then use it instead of the call to parse_crashkernel_{high|low}(). Two confusing parameters of parse_crashkernel_{high|low}() are deleted. I previously sent these four patches separately: [1] https://lkml.org/lkml/2021/12/25/40 2. Patch 0005-0009 Introduce generic reserve_crashkernel_mem[_low](), the implementation of these two functions is based on function reserve_crashkernel[_low]() in arch/x86/kernel/setup.c. There is no functional change for x86. 1) The check position of xen_pv_domain() does not change. 2) Still 1M alignment for crash kernel fixed region, when 'base' is specified. To avoid compilation problems on other architectures: patch 0004 moves the definition of global variable crashk[_low]_res from kexec_core.c to crash_core.c, and provide default definitions for all macros involved, a particular platform can redefine these macros to override the default values. 3. 0010, only one line of comment was changed. 4. 0011 1) crashk_low_res may also a valid reserved memory, should be checked in crash_is_nosave(), see arch/arm64/kernel/machine_kexec. 2) Drop memblock_mark_nomap() for crashk_low_res, because of: 2687275a5843 arm64: Force NO_BLOCK_MAPPINGS if crashkernel reservation is required 3) Also call kmemleak_ignore_phys() for crashk_low_res, because of: 85f58eb18898 arm64: kdump: Skip kmemleak scan reserved memory for kdump 5. 0012, slightly rebased, because the following patch is applied in advance. https://git.kernel.org/pub/scm/linux/kernel/git/robh/linux.git/commit/?h=dt/linus&id=8347b41748c3019157312fbe7f8a6792ae396eb7 6. 0013, no change. Others: 1. Discard add ARCH_WANT_RESERVE_CRASH_KERNEL 2. When allocating crash low memory, the start address still starts from 0. low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX); 3. Discard change (1ULL << 32) to CRASH_ADDR_LOW_MAX. 4. Ensure the check position of xen_pv_domain() have no change. 5. Except patch 0010 and 0012, all "Tested-by", "Reviewed-by", "Acked-by" are removed. 6. Update description. Changes since [v16] - Because no functional changes in this version, so add "Tested-by: Dave Kleikamp <dave.kleikamp@oracle.com>" for patch 1-9 - Add "Reviewed-by: Rob Herring <robh@kernel.org>" for patch 8 - Update patch 9 based on the review comments of Rob Herring - As Catalin Marinas's suggestion, merge the implementation of ARCH_WANT_RESERVE_CRASH_KERNEL into patch 5. Ensure that the contents of X86 and ARM64 do not overlap, and reduce unnecessary temporary differences. Changes since [v15] - Aggregate the processing of "linux,usable-memory-range" into one function. Only patch 9-10 have been updated. Changes since [v14] - Recovering the requirement that the CrashKernel memory regions on X86 only requires 1 MiB alignment. - Combine patches 5 and 6 in v14 into one. The compilation warning fixed by patch 6 was introduced by patch 5 in v14. - As with crashk_res, crashk_low_res is also processed by crash_exclude_mem_range() in patch 7. - Due to commit b261dba2fdb2 ("arm64: kdump: Remove custom linux,usable-memory-range handling") has removed the architecture-specific code, extend the property "linux,usable-memory-range" in the platform-agnostic FDT core code. See patch 9. - Discard the x86 description update in the document, because the description has been updated by commit b1f4c363666c ("Documentation: kdump: update kdump guide"). - Change "arm64" to "ARM64" in Doc. Changes since [v13] - Rebased on top of 5.11-rc5. - Introduce config CONFIG_ARCH_WANT_RESERVE_CRASH_KERNEL. Since reserve_crashkernel[_low]() implementations are quite similar on other architectures, so have CONFIG_ARCH_WANT_RESERVE_CRASH_KERNEL in arch/Kconfig and select this by X86 and ARM64. - Some minor cleanup. Changes since [v12] - Rebased on top of 5.10-rc1. - Keep CRASH_ALIGN as 16M suggested by Dave. - Drop patch "kdump: add threshold for the required memory". - Add Tested-by from John. Changes since [v11] - Rebased on top of 5.9-rc4. - Make the function reserve_crashkernel() of x86 generic. Suggested by Catalin, make the function reserve_crashkernel() of x86 generic and arm64 use the generic version to reimplement crashkernel=X. Changes since [v10] - Reimplement crashkernel=X suggested by Catalin, Many thanks to Catalin. Changes since [v9] - Patch 1 add Acked-by from Dave. - Update patch 5 according to Dave's comments. - Update chosen schema. Changes since [v8] - Reuse DT property "linux,usable-memory-range". Suggested by Rob, reuse DT property "linux,usable-memory-range" to pass the low memory region. - Fix kdump broken with ZONE_DMA reintroduced. - Update chosen schema. Changes since [v7] - Move x86 CRASH_ALIGN to 2M Suggested by Dave and do some test, move x86 CRASH_ALIGN to 2M. - Update Documentation/devicetree/bindings/chosen.txt. Add corresponding documentation to Documentation/devicetree/bindings/chosen.txt suggested by Arnd. - Add Tested-by from Jhon and pk. Changes since [v6] - Fix build errors reported by kbuild test robot. Changes since [v5] - Move reserve_crashkernel_low() into kernel/crash_core.c. - Delete crashkernel=X,high. - Modify crashkernel=X,low. If crashkernel=X,low is specified simultaneously, reserve spcified size low memory for crash kdump kernel devices firstly and then reserve memory above 4G. In addition, rename crashk_low_res as "Crash kernel (low)" for arm64, and then pass to crash dump kernel by DT property "linux,low-memory-range". - Update Documentation/admin-guide/kdump/kdump.rst. Changes since [v4] - Reimplement memblock_cap_memory_ranges for multiple ranges by Mike. Changes since [v3] - Add memblock_cap_memory_ranges back for multiple ranges. - Fix some compiling warnings. Changes since [v2] - Split patch "arm64: kdump: support reserving crashkernel above 4G" as two. Put "move reserve_crashkernel_low() into kexec_core.c" in a separate patch. Changes since [v1]: - Move common reserve_crashkernel_low() code into kernel/kexec_core.c. - Remove memblock_cap_memory_ranges() i added in v1 and implement that in fdt_enforce_memory_region(). There are at most two crash kernel regions, for two crash kernel regions case, we cap the memory range [min(regs[*].start), max(regs[*].end)] and then remove the memory range in the middle. v1: There are following issues in arm64 kdump: 1. We use crashkernel=X to reserve crashkernel below 4G, which will fail when there is no enough low memory. 2. If reserving crashkernel above 4G, in this case, crash dump kernel will boot failure because there is no low memory available for allocation. To solve these issues, change the behavior of crashkernel=X. crashkernel=X tries low allocation in DMA zone and fall back to high allocation if it fails. We can also use "crashkernel=X,high" to select a high region above DMA zone, which also tries to allocate at least 256M low memory in DMA zone automatically and "crashkernel=Y,low" can be used to allocate specified size low memory. When reserving crashkernel in high memory, some low memory is reserved for crash dump kernel devices. So there may be two regions reserved for crash dump kernel. In order to distinct from the high region and make no effect to the use of existing kexec-tools, rename the low region as "Crash kernel (low)", and pass the low region by reusing DT property "linux,usable-memory-range". We made the low memory region as the last range of "linux,usable-memory-range" to keep compatibility with existing user-space and older kdump kernels. Besides, we need to modify kexec-tools: arm64: support more than one crash kernel regions(see [1]) Another update is document about DT property 'linux,usable-memory-range': schemas: update 'linux,usable-memory-range' node schema(see [2]) [1]: https://www.spinics.net/lists/kexec/msg28226.html [2]: https://github.com/robherring/dt-schema/pull/19 [v1]: https://lkml.org/lkml/2019/4/2/1174 [v2]: https://lkml.org/lkml/2019/4/9/86 [v3]: https://lkml.org/lkml/2019/4/9/306 [v4]: https://lkml.org/lkml/2019/4/15/273 [v5]: https://lkml.org/lkml/2019/5/6/1360 [v6]: https://lkml.org/lkml/2019/8/30/142 [v7]: https://lkml.org/lkml/2019/12/23/411 [v8]: https://lkml.org/lkml/2020/5/21/213 [v9]: https://lkml.org/lkml/2020/6/28/73 [v10]: https://lkml.org/lkml/2020/7/2/1443 [v11]: https://lkml.org/lkml/2020/8/1/150 [v12]: https://lkml.org/lkml/2020/9/7/1037 [v13]: https://lkml.org/lkml/2020/10/31/34 [v14]: https://lkml.org/lkml/2021/1/30/53 [v15]: https://lkml.org/lkml/2021/10/19/1405 [v16]: https://lkml.org/lkml/2021/11/23/435 [v17]: https://lkml.org/lkml/2021/12/10/38 [v18]: https://lkml.org/lkml/2021/12/22/424 [v19]: https://lkml.org/lkml/2021/12/28/203 [v20]: https://lkml.org/lkml/2022/1/24/167 [v21]: https://lkml.org/lkml/2022/2/26/350 Chen Zhou (2): arm64: kdump: Reimplement crashkernel=X of: fdt: Add memory for devices by DT property "linux,usable-memory-range" Zhen Lei (4): kdump: return -ENOENT if required cmdline option does not exist arm64: Use insert_resource() to simplify code of: Support more than one crash kernel regions for kexec -s docs: kdump: Update the crashkernel description for arm64 .../admin-guide/kernel-parameters.txt | 9 ++- arch/arm64/kernel/machine_kexec.c | 9 ++- arch/arm64/kernel/machine_kexec_file.c | 12 +++- arch/arm64/kernel/setup.c | 17 +---- arch/arm64/mm/init.c | 64 +++++++++++++++++-- drivers/of/fdt.c | 33 +++++++--- drivers/of/kexec.c | 9 +++ kernel/crash_core.c | 3 +- 8 files changed, 118 insertions(+), 38 deletions(-)